I just made a podcast episode about my company where I work by giving it the website. It was surprisingly realistic. It also made me realize how empty many podcasts actually are.
I sent it to my colleagues telling them I "had it produced." I'll reveal the truth tomorrow.
Don't do this. A friend did this to me, and after listening to it, I suddenly realized it was AI vomit. My friend wasted an hour of my attention, and I didn't appreciate it.
If it was vomit, why did you spend an hour on it? People complain about 2 minutes of audio sometimes, I cannot imagine a full hour of an unknown podcast, it must have been quite interesting.
Because they assumed that there was a good reason that their friend sent it!?
I had a friend who did the same to me, I was sent a message asking my opinion on a tech topic. I spent 30min researching/reading to make sure my reply was accurate and then found out the question was generated by a LLM, and he just wanted to show off how good a LLM was.
It will color every interaction you have with that person...
I think you are leaving the human out of the loop. When a friend of mine recommends me something I'll lower my skepticism because I'm assuming my friend would not send me garbage.
If a random podcaster says "I've proved that P=NP" I'd say "no you didn't", but if a math professor sends me that same link I'll keep listening to see where this goes. And I've definitely read texts making wild assertions that only at the end were revealed as hit pieces and/or propaganda.
Even if you think your friend would only send good things, you would realize that something is vomit in less than an hour. I cannot understand someone listening to something for an entire hour and then whining that they waste their entire hour and it was vomit, you're not in a cinema, you didn't pay a ticket for it, you listen to something because you like it or move on.
You can argue your point all day, it will not resolve their cognitive dissonance.
No matter how convincing, high-quality or entertaining it was, no matter for how long they happily consumed the content: it's AI-generated, they hate AI, therefore it's vomit, period.
I read some of your other replies and I can't quite get a read on your line of reasoning.
The issue is we would give less attention to these things if it wasn't for the social credit the humans gave the vomit. So we engage in good faith and it turns out it was effectively a prank, and we have no choice but to value requests from those people less now because it was clear they didn't care about our response.
You ever watched a reviewbrah video? he doesn't get to "without any further ado" moment until after the halfway point of the video. The prank is the wasted time. But the joke is every other YTber does it more subversively without you getting any laughs out of it. It proves we give way more attention to slop then we dare to calculate.
I asked a friend if they had any ideas about something, and they asked an LLM, and it's like... If I wanted an LLMs answer, I'd ask it myself. I want your answer, distilled through your experience and opinions...
I don’t think this is all that impressive, the generated podcast is pretty shallow - lots of ‘whoa meta’ and the word ‘like’ thrown into every sentence.
Yes, it will generate a middle-of-the-road waffling podcast, but not one with any real depth.
I was blown away by how impressive it was. I honestly thought it was real. I still can't believe these realistic audio capabilities are not being used for pure evil everywhere we look.
> like thrown into every sentence
I think that's actually part of why it sounds real, because tons of people do actually talk like that.
To me what would make it even better is the ability to throw in random jokes and utilize information about their surroundings and recent events.
I have been using MeloTTS for text-to-speech and I thought that was about the best we could do right now, but apparently I was very wrong. Is there an offline model one can download today that sounds as good as this NotebookLM?
Bark can sound as good, but Google is using SoundStorm which was specifically trained on dialogs. Surprisingly Bark can even sort of match it without being trained to do so, but not reliably. (https://x.com/jonathanfly/status/1675987073893904386)
And SoundStorm has more than twice the context window of Bark so dialogs are a tight fit.
I just tried the default bark.cpp example from the github readme, and to me it still doesn't sound close enough to realistic, and the audio quality itself was a bit scratchy... maybe I'm doing something wrong.
When I tried my own text with it, it went completely off the rails... skipping completely over random words, and also switching to different voices in the middle of a sentence. Trying to run the large model also crashed entirely.
You aren't doing anything wrong - Bark out the box uses a randomly generated voice and I like to think it's modeling the world of random voices which includes bad microphones/audio-quality. (Even bad 'actors' - see how many Bark voices sound like they are reading a script.)
Presumably it was trained in noisy data. But it can generate and use a clean voice, they are in there. Most of the Suno default voices are not great either - but a great voice can sound perfectly clear. I haven't done much with Bark lately but on my Twitter there's plenty of clear examples of very realistic voices. Actually here I ran a prompt based on some copy and pasted test 20 times in Bark. I put a couple better results up front, but even in later samples you can find lots of evidence of human-sounding voices. https://sndup.net/bzhz5/
Going off the rails and hallucinating is a hard problem. It can be minimized, but probably would have to solved with simple brute force (check the output with S2T and retry if needed.)
For raw audio you can replace the final decoding step with something like VOCOS or MBD if you want to maximize audio quality, though you don't need do with the best voices.
This was exactly my reaction to listening to the example podcast. Although, I wonder if the base material weren't so meta-level product overview, maybe it would be better. I do think the liveliness of the conversation was good (interjections, tonal variety, etc), so at least parts of the demo are impressive.
The content is nothing that special these days, you could get it out of Gemini or Claude probably- but the audio affect is awfully convincing.
You can compare it to Google's Illuminate which also generates conversations by summarizing texts but in a much straighter, less fluffy way. It's less shallow but in some ways less compelling:
Look I agree with you at a certain level, maybe it can't emulate deep conversations about big topics (maybe it can, I haven't seen an attempt...), but a vast vast majority of podcasts and radio shows are just like this: shallow and incredibly simplified with no more than a nod to the underlying concepts. 70% personality, 20% dumb analogies that the producer thought up in thirty minutes, and <10% actually communicating the material is standard fare for normie podcasts, sadly.
Honestly, given the personalization maybe it's a net improvement.
Kind of feels like looking at an overflowing landfill and thinking "I wonder if we can invent a robot that just generates new trash directly into the landfill".
This holier than thou attitude that crops up in these threads is so annoying, as if people wanting to casually enjoy a mediocre podcast or radio show on the 1 hour commute to their shitty job is a crime.
I don’t think anyone cares about other people’s cheap pleasures. What people do care about is the displacement of quality and craft. For instance, you could say the same thing about the state of the web - say when searching for recipes. Maybe some people like the ads, the consent forms, the backstories? Why so purist? Isn’t it nice with a bit of scrolling and getting in the mood for cooking with a bit of SEO?
Defending craftsmen and attention to detail is not just about purism or gatekeeping. I appreciate people who care, even in fields I don’t personally care about (yet?). The professor who annoyingly insists on making sure every student “really gets it”, or the woodworker who is adamant about what joints are superior, or the kernel hacker who maintains rigor in face of hundreds of feature requests. The integrity of professionals can make or break institutions.
With AI reducing the effort to create garbage to the point of commoditization, people have a right, and arguably even an obligation, to be concerned. Remember, tech doesn’t follow potential, it follows incentive.
Right. Similarly, I criticize the people who worked to make cigarettes more addictive, fast food more 'craveable', freemium games more appealing to whales, gambling more attractive to problem gamblers, etc. but not people who smoke, eat fast food, play freemium games, or gamble. That would be deeply hypocritical.
I'm not criticizing the people who consume garbage, but the people who are enthusiastic about opening new markets in garbage. People should strive to do good, worthwhile things with their lives.
I think it’s “impressive” the first time you use it, but with subsequent runs it’s evident how formulaic it is. The end result, the personalities of the podcast “hosts” and their interactions are similar regardless of the context of inputs.
Basically it’s a neat party trick at the moment. I do hope to see it improve however!
Right?! We call this goalpost moving now, but it is not a new phenomena.
> It is interesting that nowadays, practically no one feels that sense of awe any longer - even when computers perform operations that are incredibly more sophisticated than those which sent thrills down spines in the early days. The once-exciting phrase "Giant Electronic Brain" remains only as a sort of "camp" cliché, a ridiculous vestige of the era of Flash Gordon and Buck Rogers. It is a bit sad that we become blasé so quickly.
> There is a related "Theorem" about progress in AI: once some mental function is programmed, people soon cease to consider it as an essential ingredient of "real thinking". The ineluctable core of intelligence is always in that next thing which hasn't yet been programmed. This "Theorem" was first proposed to me by Larry Tesler, so I call it Tesler's Theorem: "Al is whatever hasn't been done yet."
This quote is from the 80s, from GEB by Douglas Hofstadter.
(and btw, I just took a grainy, poorly-lit picture from the book, and could automagically select the text from it, since I couldn't find the quote online. Imagine that tech in the 80s. Hell, it was bad even in the 2000s, with OCR being hit and miss for a long time. Now it "just works".)
Think about how comfortable your life is, and how the 17th century version of yourself would kill to live it. Then think about how you aren't in a perpetual state of ecstasy for being given this life.
People quickly adapt to their current circumstances, take them for granted, and immediately want more.
You’re taking about advancements made through multiple lifetimes. This burst in AI has lasted about 15 years.
TBH I think it’s more of a knee jerk reaction from those tired of hearing about AI or who just want to post contrarian opinions (which I totally do sometimes, too).
To me, that's just how they tuned the 'audience' of the podcast, which I think we can imagine was at least partly informed by the 'audience' IRL podcasts are named at. I, too would like to be able to 'turn up the technical' on these, but for example, I dumped a paper about a latchless mutexless work distribution algorithm into it, which I had read but still had questions about, and the podcast accurately summarized, simplified it, and got my questions answered, which I then validated later by re-reading the paper. It was faster than combing through the paper would have been.
This is impressive from a technical point of view and probably useful from an educational one; I really like the idea that a piece of text can be transformed into any kind of media format easily, depending on your preferences. As recently as a year ago I was using Apple’s text to speech tool to listen to Wikipedia articles while biking, and needless to say, they weren’t very exciting to listen to.
But I don’t think it’s much of a threat to actual podcasts, which tend to be successful because of the personalities of the hosts and guests, and not because of the information they contain.
Which leads me to hope that the next versions of Notebook will allow more customization of the speakers’ voices, tone, education level, etc.
> But I don’t think it’s much of a threat to actual podcasts, which tend to be successful because of the personalities of the hosts and guests, and not because of the information they contain.
I wonder if any “blended” podcasts will pop up, where a human host uses a tool like this for an artificial cohost.
Latent Space AI Engineering podcast does this with an AI cohost; mostly for intros and segues. A recent episode used it to summarise a Twitter AMA and while it’s usually used to good effect, that one was one of the first episodes the quality of the co host part was lacking, as it mispronounced things, and was a bit muddled in parts.
That said, the podcast has been an incredibly useful and insightful regular listen for me.
Being able to automate words, I think, will reveal how important actual human connection is.
> We always start with a clear overview of the topic, you know, setting the stage. You’re never left wondering, “What am I even listening to?” And then from there, it’s all about maintaining a neutral stance, especially when it comes to, let’s say, potentially controversial topics.
Oh yeah, this is exactly why I listen to Oxide's podcast! (This is a joke. They often launch into topics with no explanation or context, and are unabashedly opinionated.)
AI content emulates the "production values" of high quality content, but it doesn't actually have the quality of the content it's emulating. This is why it seems impressive at first and can even fool people for quite a while. It fools our brains' heuristics for detecting good content. But when you examine it closely, the illusion falls apart. NotebookLM is not different than other generative AI products in this respect.
I do think that this will change in the not too distant future. OpenAI's o1 is a step in the direction we need to go. It will take a lot more test-time compute to produce content that has high quality to match its high production values.
I really, really hope they keep investing into NotebookLM and expand its ability to source more types of files, including codebases, complex websites etc. Feels really powerful for anybody studying or consulting many different clusters of learning materials at once.
This is amazing. I uploaded the instruction manual for a Scholander pressure chamber (a piece of equipment for measuring plant moisture stress) and made a podcast from it. The information in the podcast was accurate, it included some light banter and jokes, while still getting across the important topics in the instructions. I don't know what I would use a podcast like this for, but the fact that something like this can be created without human intervention in just a few minutes is jaw dropping, and maybe also just a teeny bit scary.
Yeah I totally get people’s criticisms that the podcasts aren’t quite human-expert-level in terms of symbolic reasoning, but this still blows my mind. The intuitive skill these show, not to mention the ability to accurately (again, if shallowly) parse and transform huge bodies of content in seconds is absolutely scary, IMO.
I’d feed it the Singularity paper, but I’m not sure I need that extra boost of anxiety these days…
We accidentally invented general models that can coherently muse about the philosophical beliefs of Gilles Deleuze at length, and accurately, based on two full books that they summarized. You can be cynical until your dying day, that’s your right — but I highly recommend letting that fact be a little bit impressive, someday. There’s no way you live through any event that’s more historically significant, other than perhaps an apocalypse or two.
In other words: soapbox is presumably some sort of toy car that goes 15mph, and formula 1 goes up above 150mph at least (as you can tell, I’m not a car guy). If you have any actual scientific argument as to why a model that can score 90-100 on a typical IQ test has only 1/10th the symbolic reasoning skills of a human, I’d love to eat my words! Maybe on some special highly iterative, deliberation-based task?
These aren't "general" models. They're statistical models. They're autocorrect or autocomplete on steroids -- and autocorrect/autocomplete don't require
symbolic reasoning.
It's also not at all clear to me what "symbolic" could mean in this context. If it means the software has concepts, my response would be that they aren't concepts of a kind that we can clearly recognize or label as such (edit: and that's to say nothing of the fact that the ability to hold concepts/symbols and understand them as concepts/symbols presupposes internal life and awareness).
The best analogy I've heard for these models is this: you take a completely, perfectly naive, ignorant man, who knows nothing, and you place him in a room, sealed off from everything. He has no knowledge of the outside world, of you, or of what your might want from him. But you slip under the door of his cell pieces of paper containing mathematical or linguistic expressions, and he learns or is somehow induced to do something with them and pass them back. When what he does with them pleases you, you reward him. Each time you do this, you reinforce a behavior.
You repeat this process, over and over. As a result, he develops habits. Training continues, and those habits become more and more precisely fitted to your expectations and intentions.
After enough time and enough training, his habits are so well formed that he seems to know what a sonnet is, how to perform derivatives and integrals, and seems to understand (and be able to explain!) concepts like positive and negative, and friend and foe. He can even write you a rap-battle libretto about nineteenth-century English historiography in the style of Thomas Paine imitating Ikkyu.
Fundamentally, though, he doesn't know what any of these tokens mean. He still doesn't know that there's an outside world. He may have ideas that are guiding his behavior, but you have no way of knowing that -- or of knowing whether they bear any resemblance to concepts or ideas you would recognize.
These models deal with tokens similarly. They don't know what a token is or represents -- or we have no reason to think they do. They're just networks of weights, relationships, and tendencies that, from a given seed and given input, generate an output, just like any program, just like your phone keyboard generates predictions about the next word you'll want to type.
Given billions and billions and billions and billions of parameters, why shouldn't such a program score highly on an IQ test or on the LSAT? Once the number of parameters available to the program reaches a certain threshold (edit: and we've programmed a way for it to connect the dots), shouldn't we be able to design it in such a way that it can compute correct answers to questions that seem to require complex, abstract reasoning, regardless of whether it has the capacity to reason? Or shouldn't we be able to give it enough data that it's able to find the relationships that enable it to simulate/generate patterns indistinguishable from real, actual reasoning?
I don't think one needs to be cynical to be unimpressed. I'm unimpressed simply because these models aren't clearly doing anything new in kind. What they're doing seems to be new, and novel, only because of the scale at which they do what they do.
Edit: Moreover, I'm hostile to the economic forces that produced these models, as everybody should be. They're the purest example of what Jaron Lanier has been warning us about -- namely that, when information is free, the wealthiest are going to be the ones who profit from it and dominate, because they'll be the ones able to pay for the technology that can exploit it.
I have no doubt Altman is aware of this. And I have no doubt that he's little better than Elizabeth Holmes, making ethical compromises and cutting legal corners, secure in the smug knowledge that he'll surely pay his mo...
I'm not saying "they predict tokens; therefore, they can't reason." I'm saying "something that can't reason can predict tokens, so prediction isn't evidence of reasoning."
More specifically, my comment aims to meet the challenge posed by the person I answered:
> I highly recommend letting that fact be a little bit impressive, someday. There’s no way you live through any event that’s more historically significant, other than perhaps an apocalypse or two. [...] If you have any actual scientific argument as to why a model that can score 90-100 on a typical IQ test has only 1/10th the symbolic reasoning skills of a human, I’d love to eat my words.
I have no idea what would constitute a "scientific argument" in this instance, given that the challenge itself is unscientific, but, regardless, the results that so impress this person are, without question, achievable without reasoning, symbolic or otherwise. To say that the model "muses" or "has [...] symbolic reasoning" is to make a wild, arbitrary leap of faith that the data, and workings of these models, do not support.
The models are token-prediction machines. That's it. They differ not in kind but in scale from the software that generates predictions in our cell-phone keyboards. The person I answered can be as impressed as he wants to be by the high quality he thinks he sees in the predictions. That's fine. I'm not. In that respect, we just disagree. But if he's impressed because he thinks the model's predictions must or do betoken reasoning, he's off in la la land -- and so his wide-eyed, bushy-tailed enthusiasm is based on nonsense.
It's no different from believing that your phone keyboard is capable of reasoning, simply because you are delighted that it guesses the 'right' word often enough to please you.
If your argument is "they can't reason" (plus some other stuff about how they work), what reasoning test has an LLM failed for you to conclude that they can't reason? Whenever I've given an LLM a reasoning test, it seems to do fine, so it really does sound to me like the argument is "they can't really reason, because <irrelevant fact about how they work internally>".
That isn't the argument. I've stated the argument twice. My longer response to you starts by stating the core of the argument as succinctly and clearly as I can. That's the first paragraph of the post. Not only are you still not getting it. You're also twisting what I wrote into claims I have not made. I'm not going to explain myself a third time.
I'll instead say this: if you think these models must be reasoning when they produce outputs that pass reasoning tests, then you should also believe, every time you see a photo of a dog on a computer screen, that a real, actual dog is somewhere inside the device.
> I'm not saying "they predict tokens; therefore, they can't reason." I'm saying "something that can't reason can predict tokens, so prediction isn't evidence of reasoning."
This is true. Reasoning is evidence of reasoning, and LLMs do pass reasoning tests. Yes, the way they work doesn't imply that they can reason, the fact that they can reason implies that.
You also said:
> These models deal with tokens similarly. They don't know what a token is or represents -- or we have no reason to think they do.
I have no reason to think that other people know what concepts are or what they represent, just that they can convincingly output a stream of words when asked to explain a context.
The argument bugs me because you can replace "they predict the next token" with "they are collections of neurons producing output voltages in response to input voltages" and you'll have the exact same argument about humans.
Thanks for explaining. I see what you're getting at now.
Here I think it's helpful to distinguish between what something is and how it's known. When we see something that resembles reasoning, we very reasonably deduce that reasoning has taken place. But 'it looks like reasoning' is not equivalent to 'it is reasoning.'
To approach the same idea from a different direction:
> I have no reason to think that other people know what concepts are or what they represent, just that they can convincingly output a stream of words when asked to explain a context.
You absolutely do have reason to think this. You're the reason. You're the best available evidence, because you have an internal life, have concepts and ideas, have intentions, and perform acts of reasoning that you experience as acts of reasoning -- and all of that takes place inside a body that, you have every reason to think, works the same way and produces the experiences the same way in other people.
So, sure, it's true that you can't prove that other people have internal lives and reason the way you do. (And you're special, after all, because you're at the center of the universe -- just like me!) But you have good reason to think they do -- and to think they do it the way you do it and experience it the way you experience it.
In the case of these models, we have no such reason/evidence. In fact, we have good reason for thinking that something other than reasoning as we think of it takes place. We have good reason, that is, to think they work just like any other program. We don't think winzip, Windows calculator, a Quake bot, or a piece of malware performs acts of reasoning. And the fact that these models appear to be reasoning tells us something about the people observing them, not about the programs themselves. These models appear to be reasoning only because the output of the model is similar enough to 'the real thing' for us to have trouble saying with certainty that they aren't the real thing. They're simulations whose fidelity is high enough to create a feeling in us -- and to pass some tests. (In that sense, they're most similar to special effects.) (Edit: and that's not to say feelings are wrong, invalid, or incorrect. They're one of the key ways we experience the things we understand.)
Is reasoning taking place in these models? Sure, it's possible. Is there an awareness or being of some kind that does the reasoning? Sure, that's possible, too. We're matter that thinks. Why couldn't a program in a computer be matter that thinks? There's a great novel by Greg Egan, Permutation City, that deals partly with this: in one section, our distant descendants pass to another universe, where matter superficially appears to be random, disorganized, and low in enthalpy. When that random activity and apparent lack of life and complexity are analyzed in the right way, though, interference patterns are revealed, and these contain something that looks like a rich vista bursting with directed, deliberate activity and life. It contains patterns that, for all the world, look and act like the universe we know -- with things that are living and things that are not, with ecosystems, predators, prey, communities, reproduction, etc. These patterns aren't in, and aren't expressed in, the matter itself. They 'exist' only in the interference patterns that ripple through it.
That's 100% plausible, too. Why couldn't an interference pattern amount to a living thing, an organism, or an ecosystem? The boundary we draw between hard, physical stuff and those patterns is arbitrary. Material stuff is just another pattern.
My point isn't that reasoning doesn't take place in these models or can't. It's, first, that you and I do something we call reasoning, and the best available information tells us these models aren't doing that. Second, if they are doing something we ...
> Indeed, we have no good reason to think there's any at all -- unless, again, we think other programs do as well.
And this is equivalent to saying there's a dog in my computer when I open a photo of a dog. It treats the simulation, the data, the program -- whatever you want to call it -- as if it were the thing itself.
Because, whenever you give it a reasoning test, it also seems to do fine.
That is what I meant in my other post, I don't really think that "it seems to do fine" is enough evidence for the extraordinary claim that it can reason.
To be fair parent pointing out they're purely statistical machines predicting next token is incorrect anyways.
They are essentially next token predictors after first training, but then instruct models are fine tuned on reasoning and Q/A scenarios, afaik early research has determined that this isn't just pure parroting, that it does actually result in some logic in there as well.
People also have to remember the training for these is super shallow at the moment, when compared with what humans go through in our lifespans as well as our millions of years of evolution (as humans).
What you're saying doesn't contradict what I wrote. I said models are trained. You're saying they're trained and fine tuned -- i.e. continue to be trained. I also didn't say they do any kind of parroting or that logic doesn't take place.
I'm saying, rather, that the models do what they're taught to do, and what they're taught to do are computations that give us a result that looks like reasoning, just the way I could use 3ds max as a teenager to generate on my computer screen an output that looked like a cube. There was never an actual cube in my computer when I did that. To say that the model is reasoning because what it does resembles reasoning is no different from saying there was an actual cube somewhere in my computer every time I rendered one.
If you had received one of these podcasts, say 3-5 years ago, I guarantee you wouldn't be able to tell it's AI generated. And I'm willing to bet it's valid for 90%+ of the people here, even those heavily involved in the field. The quality of the voices, the mimicking of umms, and ahhs, the subtle speaking over each other, they really are extremely impressive.
If you want you can do a test with people that haven't heard about the tech. Have it generate something you know they'll enjoy, maybe 2-3 min long, and have them listen to it without knowing it's AI generated. Ask them about the subject, and see if anyone mentions anything about being fake. You'll be surprised.
> 3-5 years ago, I guarantee you wouldn't be able to tell it's AI generated
You would however be able to tell that it was extremely obnoxious and bland, and without the novelty of the technical trick, you would not be listening to it.
I strongly doubt that you're gonna be listening to this stuff recreationally once the novelty wears off, but if I'm wrong and you actually enjoy listening to two robots pretend to be excited about your documents and webpages long term, then have fun with that I guess.
I'm not really a podcast listener. I've listened to maybe 20 of over the past decade, but I sometimes hear my wife listen to some. I honestly couldn't tell that this one was AI generated and it wasn't immediately obvious (for me) from the site either. So I spent a few minutes making sure it actually was AI. To me it sounded a lot like most of the English podcasts my wife has listened to, a lot of those true crime ones and it frankly could easily have been one of the tech podcast that I've listened to over the years.
I imagine it'll be even harder to know if regular pod casters feed the AI a few episode they've made, to make it learn how to talk like they do. Like, would you really know if your true crime pod casters skipped a week if the AI sounded like them? I guess I don't really fall into the category for your question as I'm not a pod cast listener.
NotebookLM's is incredibly good at generating the affect and structure of a quality podcast.
This is in-line with all art, music, and video created by LMM at the moment. They are imitating a structure and affect, the quality of the content is largely irrelevant.
I think the interesting thing is that most people don't really care, and AI is not to blame for that.
Most books published today have the affect of a book, but the author doesn't really have anything to say. Publishing a book is not about communicating ideas, but a means to something else. It's not meant to stand on its own.
The reason so much writing, podcasting, and music is vulnerable to AI disruption is that quality has already become secondary.
In the case of those books and podcasts, who cares if you read or listen to them? The point is that the books are sold and make the right lists. The point is that the podcasts are downloaded so ads can be sold or that vanity numbers can be reported.
In terms of such music and films (whether created by human or AI) sometimes it's just because we are social creatures and need shared experiences to talk with others about.
But knowing it's synthetic, why would you buy the book or listen to the podcast in the first place? There's nothing social or shared in a synthetic affectation.
In an ideal world, I would sit down with an espresso or a beer, and review collections of research papers on a regular basis.
In reality, between work, sleep and family, I rarely have anything resembling that kind of time and mental energy reserve available.
But, what I can afford is to listen to podcasts while doing other things. Doing that gives me enough of an overview to keep up with a general topic and find new topics that might be worth investing into deeper.
Wouldn’t it be great if someone made a podcast channel specifically for “Papers corysama wants to hear about at this moment”? I think so. Apparently, so do a lot of other people. But, they don’t want to list to my specific channel.
I wouldn't read an AI-generated book (except maybe once as a curiosity), but I would definitely listen to AI-generated music if it were good enough.
Reading a book is a time investment so I want it to convey the thoughts of another human being, otherwise it would feel like wasting my time. Listening to music, on the other hand, often is something that I do while I exercise, to keep a brisk pace and not get bored. As long as it sounds good, fits the genres and styles I like and is upbeat enough for exercising, I wouldn't have much of a problem with AI music - maybe it would even be a plus, since there are some specific music genres where I have already listened to pretty much everything there is (and no more is being made), and it would be great to have more.
I don't listen to podcasts, but I suppose in that case it depends on how you do so: devoting your full time and attention like a book, or as a background while you do something else like exercise music? As far as I know, many listeners are in the latter case, so I don't see why they wouldn't listen to AI podcasts.
There's background sounds and there's music. Music can communicate as much as the written word. I've listened to algorithmically generated bloop-blops and it's fine for background sound, but if it can't touch my heart it's not really music to me.
To me, as soon as I know it was fully generated it looses it's magic. It doesn't matter how good it is.
I see the same with potteries. A factory made pot cannot have more value than a hand made pot with the signature of a human. This touches the very fabric of society. Hard to explain.
I think it comes down to your area of interest. As a musician and music lover, I spend a significant amount of time trying to find or create music that is both original and good. AI generated music can be a competent imitation of well established ideas and forms, but that’s of zero value to me - I’m not looking for ‘more of the same’ - quite the opposite.
Of course. In my case, I'm not saying that I could do with AI music in any context either. Sometimes I play music in the living room, and I pay real attention to it, obviously AI won't do there. But when I'm using the music just as a background for exercising? Then sure, why not.
So you’re basically saying filler music, elevator music, backgrouound noise or whatever names it may come under. Since there’s already so much of it out there and since AI one isn’t novel in any way, I have a hard time understanding why you’d choose AI generated one.
Too much of it -> No, there are entire musical genres (e.g. italodance or big beat) where I have already listened to pretty much everything available, and they are not expanding anymore because they are not fashionable. It would be nice to have more songs and be surprised.
Isn't novel in any way -> This is not how it works, there are studies showing that AI can be creative. Or at the very least (since the definition of creativity can be controversial) produce output that is indistinguishable from novel, creative output, which is enough for the purpose discussed here.
LLMs are already better than books for exploring some ideas. But in conversation form.
Until we get better versions of o1 that can generate insights over days and then communicate them in book form the loss of interactivity and personalisation makes LLM books pointless.
An interactive conversation / tutorial session beats a book pretty much all of the time. Nonfiction books contain a lot of information that's redundant to a reader familiar with the topic, and not enough for someone new. They don't backtrack if you clearly missed an important point. And so on. It's like fractal geometry.
If an AI agent understands you (and book writing, and the topic of the book) well enough then it should be able to write you a pretty nice bespoke book.
I do suspect that interactive media is just strictly better in theory. But maybe there will be a period of time where bespoke AI-generated books make sense.
The problem is a whole book worth is a long time to go between feedback and questions. I don't see how the agent would know the reader that well, knowledge is embedded in the brain and only comes out when prompted.
Your comment implies that there is an existing piece of music, which can subsitute the generated music. While subsitutability varies from person to person, your original statement implies for me that each generated music has an accompanying original music that you can listen to instead (of which it was "stolen" from), since it is similar enough. I think we both know that that is not the case.
I know that you likely intended to imply that you can subsitute the aformentioned AI music with an existing piece of music of the same genre, but that is not a view shared by all. Sometimes the generated music scratches such a specific and personal itch, that it cannot be replicated by something in the same genre.
A better counterargument to your original comment would be "It is not an exclusive situation. I can listen to and support both generated music and handcrafted music at the same time. They both contain music tracks that I like."
You don't have to be a big fucking nerd about it, you know what I meant. The generated music wouldn't exist without the foundation of stolen music made by people.
No, I didn't know what you meant. Communication is hard, and there are multiple ways to interpret your statements. It is better to be specific.
To be more specific about the second sentence, if there are any readers in doubt:
> The generated music wouldn't exist without the foundation of stolen music made by people.
The word "stolen" is a value judgement that is not shared by all. It is a word meant to invoke an emotional response in the reader. For example, Stallman has argued that the data could not have been stolen, or else it would not be there anymore. So, removing this word gives you:
> The generated music wouldn't exist without the foundation of existing music made by people.
Which is a true fact that has never been in debate.
However, this is not relevant to the main point that not all generated music has a suitable handcrafted substitute, and that there is no actual need to choose exclusively to listen to generated or human crafted music. Furthermore, the conversation has turned uncivil (the first sentence). Therefore, goodbye.
>But why would I buy those books or listen to those podcasts that are synthetic affectations of no substance?
A randomly selected NotebookLM podcast is probably not substantial enough on its own. But with human curation, a carefully prompted and cherry-picked NotebookLM podcast could be pretty good.
Or without curation, I would use this on a long drive where audio was the only option to get a quick survey of a bunch of material.
That's the same question I have. There is already a ton of great podcasts/music/everything in the niches that I like that I don't have the time to listen to them all. I also like to have quiet introspective time.
So where does AI regurgitated slop fit into my life?
In the case of NotebookLM, the AI generated podcasts aren't competing with existing podcasts, they're competing with other ways of consuming the source material. Would I rather listen to a real podcast? Yes. But no one's making a real podcast about the Bluetooth L2CAP specification.
I think and hope that you're wrong. There's always been cheese, and there's a lot of it now. But there is still a market for top-notch insight.
For example, Perun. This guy delivers an hourlong presentation on (mostly) the Ukraine-Russia war and its pure quality. Insights, humour, excellent delivery, from what seems to be a military-focused economist/analyst/consultant. We're a while away from some bot taking this kind of thing over.
Seriously, hardcore history? I dont even remember where I heard from him, but I think it was a Lex podcast. So I checked out hardcore history and was mightily disappointed. To my ears, he is rambling 3 hours about a topic, more or less unstructured and very long-winded, so that I basically remember nothing after having finished the podcast. I tried several times again, because I wanted it to be good. But no, not the format for me, and not a presentation I can actually absorb.
Yea there are much better examples of quality history podcasts, that are non-rambling. E.g. Mike Duncan podcasts (Revolutions, History of Rome), or the Age of Napoleon podcast. But even those are really just very good digestions of various source materials, which seems like something where LLMs will eventually reach quite a good level.
For such historical topics, my LLM-based software podgenai does a pretty good job imho. It is easier for it since it's all internal knowledge that it already knows about.
It's interesting I have the exact opposite opinion. I'm sure Mike Duncan works very hard, and does a ton of research, and his skill is beyond anything I can do. But his podcasts ultimately sound like a list of bullet points being read off a Google Doc. There's no color, personality, or feeling. I might as well have a screen reader narrate a Wikipedia article to me. I can barely remember anything I heard by him.
Carlin on the other hand, despite the digressions and rambling, manages to keep you engaged and really feel the events.
Hardcore History can certainly be off kilter, and the first eppy of any series tends to be a slog as he finds his groove. That said, Wrath of the Khans, Fall of the Republic, and the WW1 series do blossom into being incredible gripping series.
I know I'm not the first to say this, but I think what's going on is that these AI things can produce results that are very mid. A sort of extra medium. Experts beat modern LLMs but modern llms are better than a gap.
If you just need voice discussing some topic because that has utility and you can't afford a pair of podcasters (damn, check your couch cushions) then having a mid podcast is better than having no podcast. But if you need expert Insight because expert Insight is your product and you happen to deliver it through a podcast then you need an expert.
If I were a small software shop and I wanted something like a weekly update describing this week's updates for my customers and I have a dozen developers and none of us are particularly vocally charismatic putting a weekly update generated from commits, completed tickets, and developer notes might be useful. The audience would be very targeted and the podcast wouldn't be my main product, but there's no way I'd be able to afford expert level podcasters for such a position.
I would argue Perun is a world class defense Logistics expert or at least expert enough, passionate enough, and charismatic enough to present as such. Just like the guys who do Knowledge Fight, are world class experts on debunking Alex Jones, and Jack Rhysider is an expert and Fanboy of computer security so Darknet Diaries excels, and so on...
These aren't for making products, they can't compete with the experts in the attention economy. But they can fill gaps and if you need audio delivery of something about your product this might be really good.
Edit - but as you said the robots will catch up, I just don't know if they'll catch up with this batch of algorithms or if it'll be the next round.
> I know I'm not the first to say this, but I think what's going on is that these AI things can produce results that are very mid. A sort of extra medium. Experts beat modern LLMs but modern llms are better than a gap.
I've seen people manage to wrangle tools like Midjourney to get results that surpass extra medium. And most human artists barely manage to reach medium quality too.
The real danger of AI is that, as a society, we need a lot of people who will never be anything but mediocre still going for it, so we can end up with a few who do manage to reach excellence. If AI causes people to just give up even trying and just hit generate on a podcast or image generator, than that is going to be a big problem in the long run. Or not, and we just end up being stuck in world that is even more mediocre than it is now.
> as a society, we need a lot of people who will never be anything but mediocre still going for it, so we can end up with a few who do manage to reach excellence
"Reach excellence" is the key phrase there. Excellence takes time and work, and most everyone who gets there is mediocre for a while first.
I guess if AIs become excellent at everything, and the gains are shared, and the human race is liberated into a post-scarcity future of gay space communism, then it's fine. But that's not where it's looked like we're heading so far, though - at least in creative fields. I'd include - perhaps not quite yet, but it's close - development in that category. How many on this board started out writing mid-level CRUD apps for a mid-level living? If that path is closed to future devs, how does anyone level up?
> But that's not where it's looked like we're heading so far
I think one of the major reasons this is the case is because people think it's just not possible; that the way we've done things is the only possible way we can continue to do things. I hope that changes, because I do believe AI will continue to improve and displace jobs.
My skepticism is not (necessarily) based on the potential capabilities of future AI, it's about the distribution of the returns from improved productivity. That's a political - not a technological - problem, and the last half century has demonstrated most countries unable to distribute resources in ways which trend towards post-scarcity.
That may be your position as well - indeed, I think your point about "people think[ing] it's not possible" is directly relevant - but I wanted to make that more explicit than I did in my original comment.
AI looks like it will commoditise intellectual excellence. It is hard to see how that would end up making the world more mediocre.
It'd be like the ancient Romans speculating that cars will make us less fit and therefore cities will be less impressive because we can't lift as much. That isn't at all how it played out, we just build cities with machines too and need a lot less workers in construction.
If you want to say AI have reached intellectual Excellence because we have a few that have peaked in specific topics I would argue that those are so custom and bespoke that they are primarily a reflection on their human creators. Things like Champions and specific games or solutions to specific hard algorithms are not generally repurposable, and all of the general AI we have are a little bit dumb and when they work well they produce results that are generally mid. On occasionally we can get a few things we can sneak by and say they're better but that's hardly a commodity that's people sifting through large piles of mid for gems.
There are a lot of ways if it did reach intellectual excellence that we could argue that it would make Humanity more mediocre, I'm not sure I buy such arguments but there are lots of them and I can't say they're all categorically wrong.
> It'd be like the ancient Romans speculating that cars will make us less fit and therefore cities will be less impressive because we can't lift as much. That isn't at all how it played out
No, obviously not. Modern construction is leagues outside what the Romans could ever hope to achieve. Something like the Burj Khalifa would be the subject of myth and legend to them.
We move orders of magnitude more cargo and material than them because fitness isn't the limiting factor on how much work gets done. They didn't understand that having humans doing all that labour is a mistake and the correct approach is to use machines.
I don't know, Dubai is...bigger, but I'd say it's vastly more mediocre city than Rome. To your original point, making things easier to make probably does exert downward pressure on quality in the aesthetic/artistic sense. Dubai might have taller buildings and better sewage system[0], but it will never have the soul of a place like Rome.
[0] Given the floods I saw recently, I'm not even sure this is even true.
I don't think you're logic follows that we need a lot of people suffering to get a few people to be excellent. If people with a true and deep passion follow a thing I think they have a significant chance of becoming excellent at it. These are people who are more likely to try again if they fail, these are people who are more likely to invest above average levels of resources into acquiring the skill, these are people who are willing to try hard and self-educate, such people don't follow a long tail distribution for failure.
If someone wants to click generate on a podcast button or image generator it seems unlikely to me that was a person who would have been sufficiently motivated to make an excellent podcast or image. On the flip side, consider if the person who wants to click the podcast or image button wants to go on to do script writing, game development, Structural Engineering, anything else but they need a podcast or image. Having such a button frees up their time.
Of course this is all just rhetorical and occasionally someone is pressed into a field where they excel and become a field leader. I would argue that is far less common than someone succeeding and I think they want to do, but I can't present evidence that's very strong for this.
This is true but the quality frontier is not a single bar. For mainstream content the bar is high. For super-niche content, I wouldn’t be surprised if NotebookLM already competes with the existing pods.
This will be the dynamic of generated art as it improves; the ease of use will benefit creators at the fringe.
I bet we see a successful Harry Potter fanfic fully generated before we see a AAA Avengers movie or similar. (Also, extrapolating, RIP copyright.)
On the contrary, the mainstream eats any slop you put infront of it as long as it follows the correct form - one needs only look at cable news - the super niche content is that which requires deep thinking and novel insights.
Or to put another way, I've heard much better ideas on a podcast made by undergrad CS students than on Lex Fridman.
It's the complete opposite. Unless your definition of mainstream includes stuff like this deep drive into Russia/Ukraine, in which case I think you're misunderstanding "mainstream".
We all seek different kinds of quality; I don't find Peruns videos to have any quality except volume. He reads bullet points he has prepared, and makes predictable dad jokes in monotone, re-uses and reruns the same points, icons, slides, etc. Just personally, I find it really samey and some of the reporting has been delayed so much it's entirely detached from the ground by the time he releases. It's a format that allows converting dense information and theory to hour long videos, without examples or intrigue.
Personally, I prefer watching analysis/sitrep updates with the geolocations/clips from the front/strategic analysis which uses more of a presentation (e.g. using icons well and sparingly). Going through several clips from the front and reasoning about offensives, reasons, and locations is seems equally difficult to replicate as Peruns videos, which rely on information density.
I do however love Hardcore history - he adds emotion and intrigue!
I agree with your overall hope for quality and different approaches still remaining stand out from AI generated alternatives.
I like a range of the Ukraine coverage. From stuff that comes in fast to the weekly roundup-with-analysis. E.g. Suchomimus has his own humour and angle on things, but if you don’t have a unique sense of humour or delivery then it’s easier for an AI to replace you.
Give it a year or three, up to the minute AI generated sitrep pulling in related media clips and adding commentary…not that hard to imagine.
> Give it a year or three, up to the minute AI generated sitrep pulling in related media clips and adding commentary…not that hard to imagine.
But why? Isn’t there enough content generated by humans? As a tool of research AI is great in helping people do whatever they do but having that automated away generating content by itself is next to trash in my book, pure waste. Just like unsolicited pamphlets thrown at your door you pick up in the morning to throw in the bin. Pure waste.
All of the other commentators have replied with a good diverse set of YouTubers and included ones with biases from both sides; I'd recommend the ones they have linked. Some (take note of the ones that release information quicker) might be more biased or more prone to reporting murky information than others.
I think the main problem with Peruns' videos are that they are videos. I run a little program on my home-lab that turns them into podcasts and I find that I enjoy them far more because I need to be less engaged with a podcast to still find them enjoyable. (Also, I gave up on being up to date with Ukraine situation, since up to date information is almost always wrong. I am happy to be a week or a 14 days behind if the information I am getting is less wrong).
I like Hardcore history very much, but I think it would be far worse in a video form.
> He reads bullet points he has prepared, and makes predictable dad jokes in monotone, re-uses and reruns the same points, icons, slides, etc.
The presentation is a matter of taste (I like it better than you do), but the content is very informative and insightful.
Its not really about what is happening at the frontline right now. Its not its aim. Its for people who want dense information and analysis. The state of the Ukrainian and Russian economies (subjects of recent Perun videos) does not change daily or weekly.
I would like them to be right, for that to mean that the 'real' content gets fewer (fewer bother) but better (or at least higher SNR among what there is).
And then faster/easier/cheaper access to the LM 'uninspired but possibly useful' content, whatever that might look like.
I keep seeing this asertion: "the robots will get there" (or its ilk), and it's starting to feel really weird to me.
It's an article of faith -- we don't KNOW that they're going to get there. They're going to get better, almost certainly, but how much? How much gas is left in the tank for this technique?
Honestly, I think the fact that every new "groundbreaking" news release about LLMs has come alongside a swath of discussion about how it doesn't actually live up to the hype, that it achieves a solid "mid" and stops there, I think this means it's more likely that the robots AREN'T going to get there some day. (Well, not unless there's another breakthrough AI technique.)
Either way, I still think it's interesting that there's this article of faith a lot of us have "we're not there now, but we'll get there soon" that we don't really address, and it really colors the discussion a certain way.
AI has been so conflated with LLMs as of late that I'm not surprised that it feels like we won't get there. But think of it this way, with all of the resources pouring into AI right now (the bulk going towards LLMs though), the people doing non-LLM research, while still getting scraps, have a lot more scraps to work with! Even better, they can probably work in peace, since LLMs are the ones under the spotlight right now haha
IMO it seems almost epistemologically impossible that LLM's following anything even resembling the current techniques will ever be able to comfortably out-perform humans at genuinely creative endeavours because they, almost by definition, cannot be "exceptional".
If you think about how an LLM works, it's effectively going "given a certain input, what is the statistically average output that I should provide, given my training corpus".
The thing is, humans are remarkably shit at understanding just how exception someone needs to be to be genuinely creative in a way that most humans would consider "artistic"... You're talking 1/1000 people AT best.
This creates a kind of devils bargain for LLMs where you have to start trading training set size for training set quality, because there's a remarkably small amount of genuinely GREAT quality content to feed this things.
I DO believe that the current field of LLM/LXM's will get much better at a lot of stuff, and my god anyone below the top 10-15% of their particular field is going to be in a LOT of trouble, but unless you can train models SOLELY on the input of exceptionally high performing people (which I fundamentally believe there is simply not enough content in existence to do), the models almost by definition will not be able to outperform those high performing people.
Will they be able to do the intellectual work of the average person? Yeah absolutely. Will they be able to do it probably 100/1000x faster than any human (no matter how exceptional)?... Yeah probably... But I don't believe they'll be able to do it better than the truly exceptional people.
Yes, LLMs are probably inherently limited, but the AI field in general is not necessarily limited, and possibly has the potential to be more genuinely creative than even most exceptional creative humans.
I loosely suspect too many people are jumping into LLMs and I assume real research is being strangled. But to be honest all of the practical things I have seen such as by Mr Goertzel are painfully complex very few can really get into.
I’m not sure. The bestsellers lists are full of average-or-slightly-above-average wordsmiths with a good idea, the time and stamina to write a novel and risk it failing, someone who was willing to take a chance on them, and a bit of luck. The majority of human creative output is not exceptional.
A decent LLM can just keep going. Time and stamina are effectively unlimited, and an LLM can just keep rolling its 100 dice until they all come up sixes.
Or an author can just input their ideas and have an LLM do the boring bit of actually putting the words on the paper.
I’m just saying, the vast majority of human creative endeavours are not exceptional. The bar for AI is not Tolkien or Dickens, it’s Grisham and Clancy.
IMO the problem facing us is not that computers will directly outperform people on the quality of what they produce, but that they will be used to generate an enormous quantity of inferior crap that is just good enough that filtering it out is impossible.
We have already trashed the internet and really human communication with SEO blogspam brought even lower by influencers desperately scrambling for their two minutes of attention. I could actually see quality on average rising, since it will now be easy to churn out higher quality content, even more easily than the word salad I have been wading through for at least the last 15 years.
I am not saying it's not a sad state of affairs. I am just saying we have been there for a while and the floor might be raised, a bit at least.
Agreed. I think people are extrapolating with a linearity bias. I find it far more plausible that the rate of improvement is not constant, but instead a function of the remaining gap between humans and AI, which means that diminishing returns are right around the corner.
There's still much to be done re: reorganizing how we behave such that we can reap the benefits of such a competent helper, but I don't think we'll be handing the reigns over any time soon.
In addition to "will the robots get there?" there's also the question "at what cost?". The faith-basedness of it is almost fractal:
- "Given this thing I saw a computer program do, clearly we'll have intelligent AI real soon now."
- "If we generate sufficiently smart AI then clearly all the jobs will go away because the AI will just do them all for us"
- "We'll clearly be able to do the AI thing using a reasonable amount of electricity"
None of these ideas are "clear", and they're all based on some "futurist faith" crap. Let's say Microsoft does succeed (likely at collosal cost in compute) in creating some humanlike AI. How will they put it to work? What incentives could you offer such a creature? What will it want in exchange for labor? What will it enjoy? What will it dislike? But we're not there yet, first show me the intelligent AI then we can discuss the rest.
What's really disturbing about this is hype is precisely that this technology is so computationally intensive. So of course the computer people are going to hype it--they're pick and shovel salespeople supplying (yet another) gold rush.
I stumbled on a parody of Dan Carlin recently. I don't know the original content enough to know if it's accurate or even funny as a satire of him specifically, but I enjoyed the surreal aspect. I'm guessing some AI was involved in making it:
I thought this was a great, insightful comment, but noodling over it a little more made me think it's not just content producers who are responsible for this "quality vacuousness" epidemic.
I think this is just partly an inevitable consequence of going from "content scarcity" to our new normal of "content obesity" over the past 20 years or so. In this new era of an overwhelming amount of content, it's just natural to compare it all against each other, e.g. to essentially "optimize" it to the "best" form, but in doing that we've fallen into a homogeneity, and the resulting lack of variation is an actual lowering of quality in and of itself.
2 examples to explain what I mean:
1. I find that nearly all interior design (at least within broad styles) looks basically the same to me now. It's all got that "minimalist, muted tones but with a touch of organic coziness and one or two pops of color" look to it. Honestly, I don't know how interior designers even exist today, when it's trivial to go to Houzz or any of a million websites and say "yes, like this". A while back I was complaining online somewhere that I thought all interior design looked similar where in the past there was much more interesting variation, and somebody insightful replied that it's not really that interior design is now just the same, it's that it's really just converged. People can easily see and compare a million designs against each other, so there is much less of a chance for that green shag carpet to even get a moment in the sun.
2. I was recently on vacation and decided I wanted to read a "classic" book, so I read Hemingway's The Sun Also Rises (I'm not sure why I never had to read that in high school). Nearly throughout the entire book I couldn't help but thinking "Is there any time this book stops sucking?" I hated the entire thing - it was like being forced to watch someone's vacation photos for twelve hours straight, and I kept wondering why there never seemed to be any attempt to actually make me give a shit about any of the characters in the book, as nearly every one of them I found insufferable and wondered how they each had about 3 or 4 livers to spare. But I do understand that Hemingway's writing style was unique and original at the time, and that he was doing something new and interesting that influenced American literature for a long time. But these days, given the flood of content, it feels like most attempts at doing something "new and interesting" are not only forced, but nearly impossible given that there are a million other people also trying to do new and interesting things that now have the means to disseminate them. I don't think a book like The Sun Also Rises, where I believe the main impact was the style of writing/dialogue vs the actual story, could ever break through today.
I guess my point with this long post is that I think the "loss of quality" in content that many of us sense is just a direct result of there being so much content that we see variations from the "ideal" as worse, where in the past we may have found them interesting.
You're right, I love this, thanks! I was familiar with some of these examples, e.g. Komar and Melamid's painting example (and, IIRC, unless I'm confusing with other artists, they also painted a painting filled with features that the "average" person hated, like abstract geometric shapes and stark colors, and the artists actually liked that painting and said something along the lines of "turns out we're really good at making bad art"), and the "AirBnB-style of interior design" was so excellently skewered by SNL recently, and HN has had a number of posts about how so many brands have devolved to the same monochrome, san-serif typefaces for their logos.
Still, at the same time, I couldn't help but feeling a little bit sad/resigned at the existence of the article you linked. Here I thought I had an idea that was not exactly unique but that I felt would be good to share. And yet then here is an example that expresses this idea a million times better than I ever could (I love "The Age of Average" headline), with great researched examples and tons of helpful visuals. It's hard to not feel a bit like Butters in that "Simpsons did it!" episode of South Park...
What you say (though I'm not sure that we can speak of an "ideal"), compounded with the "late stage capitalism" fact that everything today is consolidated, and has to be about making profit and maximizing it: Disney shareholders probably like the latest Marvel movie more than you do for being the same as the previous ones: business don't like taking risks. The same applies to your furniture maker: when you sell to millions and want your shelves stuffed, you pick a select few materials and color variations that minimize cost and targets the broadest audience.
I ran one of my papers into it, mind blown how well they dumbed it down without losing too much details (still quite a lot was ommitted). I wonder if it's domain specific, and I wonder what's the variance by topic.
Same here. In fact, I typically struggle communicating my scientific research to journalists, and next time I'll use this. It found some good metaphors to make even a quite math-heavy paper's core concepts understandable to the audience without losing correctness, which is something that both I and the journalist typically fail to do (I keep the correctness but don't make it understandable enough, so then journalists start coming up with metaphors and do the opposite).
A lawyer friend of mine also suggested giving it the Spanish civil code, a long, arid legal text. The podcast of course didn't cover the whole text in 10 minutes, which would be impossible, but they selected some interesting tidbits and actually had me hooked until the end and made me learn a few things about it, which is no small merit. And my friend was quite impressed and didn't complain about correctness.
I did the same thing, running one book I edited and another book I wrote through it, and it did quite well. I was particularly impressed with how the “hosts” came up with their own succinct examples and metaphors to explain what I had written at much greater length. (I should mention that one of those books was in Japanese, and they captured it clearly in English.)
Lately, when I just want to get the gist of a long article or research paper, I run it through NotebookLM and listen to the podcast while I’m exercising.
My only complaint is that the chatty podcasty gab gets tiring after a while. I wish it were possible to dial that down.
I think it is right that people don't care and there is some merit to it.
Reading, or listening to podcast, these days is more akin to a meditation - many people do it to reenforce an identity rather than to expand on themselves.
And I do think that is reasonable as, for many people, there are few other structures that can keep them in check with themselves.
I would disagree it's trying to be a "quality" podcast. As usual with AI, it's an average over averages, incredibly mediocre, sometimes borderline satire. For instance, in this example podcast they say "and trust me, guys, you wanna hear all about this", which is where I would usually turn off, because nothing of quality can come after this sentence.
In my company, HR now uses AI to do training videos. It's hilariously funny, because it looks like a satire on training videos (well, granted, it's funny for a minute or two, then it shifts to annoying).
That's actually a really good application of AI, because the quality of the content is meaningless as long as it hits the bullet points. They only do this to check a box that training on <topic> was done.
Right? The fact that the LLM output is indistinguishable from a podcast says more about podcasts than about LLMs.
If anything, listening to that reminded me of why I stopped listening to podcasts in the first place - every 5 second snippet of something interesting ends up suffocated by 5 minutes of filler and dead air.
So your argument in anutshell is: humans have nothing to say, let's stop listening to them. Are you serious? It's ALL about what humans want to send out to the world, this is what it's all about. I'm perplexed that this isn't obvious.
Think back to the mid-1980s and the first time everyone got their hands on a Casio or Yamaha keyboard with auto-accompaniment.
It was a huge amount of fun to play with, just pressing a few buttons, playing a few notes and feeling like you were producing a "real" pop song. Meanwhile, any actual musicians were to be found crying in the corner of the room, not because a new tool had come along which threatened their position, but because non-musicians apparently didn't understand (at least immediately) the difference between these superficial, low-effort machine-generated sounds and actual music.
To be clear, Dan the Automator added an additional drum track, an additional bass track, and a melodica track as well as numerous other sound effects. They didn't just loop the Casio demo track.
What is scary about AI is the speed of improvement, not what it currently is.
People keep forming these analogies/explanations with the inherent premise that what we have now is what AI is going to be - "It's actually kind of shitty so don't fret, not much will change".
AI music creation has improved more in the last 5 years than keyboard accompaniment improved in the previous 40 years. It would be very brazen to bet that the tech 5 years from now is hardly any better. Especially when scaling transformers has consistently improved outputs. Double especially when the entire tech industry is throwing the house at scaling it.
Popular music has already been synthetic and souless for decades now. People will listen to what sounds good to them, and we already know the bar is very low, and that the hard truth is that it is all subjective anyway.
More of a behavioural science take. Is music the sound that is played or the people making the sound?
We’ve had software accompaniment for a long time. Elevator music. The same 4 chords arranged in similar ways for decades. Hasn’t destroyed music. Neither will AI.
At some point people are going to want to know who’s on the other side making the music.
Unless your argument is that nobody values artists… which is I guess one of the primary conceits of GenAI enthusiasts today.
Remembering the 90s when I grew up really into alternative music I think what has changed is maybe public perception. AI back then would not have changed much because mainstream pop music was already accepted as generic derivative existing only to make money. Quality was already seen as secondary to be successful. But nowadays maybe due to social networks incentives instead of journalists curation only numbers seem to matter.
Well, yes. Replace the various music and book publishing mills with LLMs for even more low quality drivel filling the marketplaces because now even the already low barrier of having to actually pay someone to produce it will be removed.
That's definitely going to be an improvement. Not.
> The reason so much writing, podcasting, and music is vulnerable to AI disruption is that quality has already become secondary.
I think that has always been the case, we just tend to compare today’s average stuff with the best stuff from earlier days.
For example, most furniture pictures from the 60s and 70s are from upper middle class homes. If we listen music, we listen to Queen and not some local band from Alabama (not that I’m against such bands at all; they can make great music too).
> I think that has always been the case, we just tend to compare today’s average stuff with the best stuff from earlier days.
I agree with this of course, because generally nobody remembers the bad stuff unless it was the worst. I beg to differ with music, though, because there's an opposing effect: we tend to be left with the most marketed music, which was usually a cheap knockoff of something interesting going on at the time. The shitty commercial knockoff becomes the "classic" while the people they were ripping off don't even get a wikipedia page.
You're raising a good point about how "best" is defined.
If you ask most people, they are by definition more likely to connect with broadly disseminated cheap knock offs than they are with whatever 'legit' inventive underground creator, simply because they've heard the former and not the latter.
Just a mental exercise: If you ask 1000 people if they prefer Knock Off or Original, and 900 say Knock Off, which one was better? If the answer is still Original, by what metric do we measure quality?
it is the perfect milquetoast personality. It's like don lemon but without the interesting bits of don lemon. It has no draw or interest.
Podcasts are only somewhat about things. The most important part is that they're by people, and the people is what draws people in. These ai podcasts are not by people, and when you listen to more than one you start to see the patterns and void where a personality is.
Yes, this is impressive, it has all the idiosyncrasies of podcasting, the pauses, turns of phrase, even the tones where we hear people putting things in quotes, etc.
... but it's also pointless. And it's likely different episodes on different topics will tend to sound very much alike; it's already the case here, I'm sure I heard another example where the two voices were the same.
In less than a year we all have learned to recognize AI images with pretty good accuracy; text is more difficult, but podcasting seems easy in comparison.
yes, podcasting is a goto market strategy. One reason there are so many VC podcasts is because it is how GPs (VCs who fundraise) reach LPs (the money that invests in venture funds).
I think the average person is more interested in the output than in the process e.g. more people want to read The Shining than want to read about how The Shining was written
We’ve become so great at articulation and delivery of empty ideas. To a point, I completely block out people like these in real life. This is an entire career for many.
my first job out of college was at a big name management consulting firm... to riff on your point: yes, such is the entire career for many. and theirs aren't even such bad careers if one only considers money and prestige. two years there completely cured me of any illusion of positive correlation between prestige and intelligence. I used to wonder if the partners at the firm actually believed the bullshit they were spilling -- actually "delivering value" per consulting parlance. I get it that people do intellectually dishonest things just for the money... but the partners seemed to genuinely believe their chatgpt-esque text generation. In the end I figured it was a combination of self-selection (only the true believers stay for the years and make partner) and a psycho-hack where if you want to convince your client, you better believe it yourself first (only the true believers make good evangelists).
It ends up sounding like a smarmy Sunday-morning talk show conversation, with over-exaggerated affect and no content.
So far I've just fed it technical papers, which may be part of the problem, but what I got back was, "Gosh, imagine if a recommender system really understood us? Wow, that would be fantastic, wouldn't it?"
While it's impressive, I agree that it tends to make over the top comments or reactions about everything. It could probably make a Keurig machine sound like a revolutionary coffee maker.
Already in the sample embedded by Simon. "Gosh", "wow", "like", "like", "like", "[wooooaaaawiiiing, woooooooaawiiiiiiing]", "Oh my god", "I was so, like...".
> The reason so much writing, podcasting, and music is vulnerable to AI disruption is that quality has already become secondary.
Commercial creative workers are vulnerable because there's a billions-of-dollars industry effort to copy their professional output and compete with them selling cheap knock-offs.
I see this sort of convenient resignation all the time in the tech crowd... "creative workers only can blame themselves for tech companies taking their income because their art just isn't any good anymore!"
The poor quality "content" that's been proliferating recently has been created, largely, using the very tools that AI has built, or their immediate precursors. AI, for all its benefits, has only made that worse.
If you're saying, in good faith, that most of the infomercials, televangelist programs, talk radio, celebrity autobiographies, self-help books, scandalous expose books, and health/exercise fad books etc etc etc that came out 50 years ago were made for no reason beyond advancing human knowledge, you're either too young to remember any media from before our current era and haven't looked beyond survivorship bias.
Tech folks love sentiments like this because it entirely emotionally places the onus on the people getting ripped off by big tech companies for being ripped off. If their work was that awful, companies wouldn't be clamoring to vacuum it up into their models to make more of it. Nearly all of the salable output from these models exists solely because it took a creative product someone made with the intention of selling it and it's using it to sell a simulacra.
It's using nostalgia to deflect guilt for harpooning the livelihood of many people because it's just more convenient and profitable to empower mediocre "content creators" they use to justify doing it.
I'm confused about your point RE: infomercials et al - that's poor quality "content" that's been proliferating for, as you say, more than 50 years.
Is that not the work of commercial creative workers? Did it not exist pre AI? There's an argument to scale, certainly, but the idea that "things were better in the past before these <<new technologies>> came out" is generally a suspect argument.
To your broader point - new tools for creating creative work come out all the time. Did we suffer greatly at the loss of image compositors when Photoshop arrived? On the flip side, did digital art gut painting and sculpture? Isn't this just another tool for creative expression?
Art is a way of seeing, not a way of creating. I don't think the technology is taking that away.
>Is that not the work of commercial creative workers? Did it not exist pre AI? There's an argument to scale, certainly, but the idea that "things were better in the past before these <<new technologies>> came out" is generally a suspect argument.
The fact that all of that stuff was crap is central to my point. You might just need to give it another read.
> Art is a way of seeing, not a way of creating. I don't think the technology is taking that away.
I'm really sick and tired of the tech industry's bumper-sticker-level-reductive pseudo-philosophical generalizations about "what art is," what it means to be an artist, the acceptable ways to be an artist, and all of that. Art is a whole fucking lot of things, and chief among them in this context is a class of professions. Glib decrees based on a razor-thin slice of one of the broadest topics in the human experience that conveniently exclude or dismiss the stakes of those with the loudest criticism and the most to lose is obviously self-serving. If you're going to take the libertarian "well that's the market for ya" stance," at least be honest about it. If you're going to try to carefully define the entire universe of ideas and practices that comprise art to conveniently exclude the concerns of the people getting screwed over because you think the optics are better or you feel less icky about it, well you better expect to get some really pissed off responses from them.
There's no glib decree - a new technology has arrived. I'm being very honest about it - you can't put it back in the bottle, any more than you could put jacquard looms or machine woodcarving of trim could be. The position between art and craft can be endlessly debated and I put my stake in the ground.
You can disagree! Folks who are impacted have every right to be pissed, organize, take action. All of these creative endeavors existed _post technology updates_ though - that's my entire point. The need for art doesn't disappear - it changes. Standing athwart the change is a choice, but I'm not sure it is an effective position.
Well gosh, good thing someone in big tech gave me permission to be mad about many in my field being screwed by big tech! Too bad that won't help pay for my cancer treatment because there's no way in hell they'll push out a cure soon enough when they're dumping billions of dollars into figuring out how to sell other people's artwork. At least people won't have to waste an uncomfortable few minutes writing a thoughtful note to my wife in the aftermath when they can just "Ok, google" it.
>> Art is a way of seeing, not a way of creating.
> There's no glib decree
This is a glib decree and it completely ignores most of what art actually is in our world, rather than the quaint little box that most people in the NN business try to stuff it into. Your patronizing tone doesn't lend any authority or add depth to your initial analysis, which you essentially just restated using more words. The "art vs craft" dichotomy doesn't even approach the depth and complexity of the interplay of art and commerce in the worlds like video game development, music, cinema and television, and writing... hell even advertising. Like most tech dudes that assume their incredible mental might gives them some kind of pan-topic expertise allowing them to casually dismiss subject matter experts in other fields based on a few a priori thought exercises, you simply don't know how much more you need to learn to make informed decisions about this topic.
Sure, and it wasn't the right thing to do, especially if it was from an independent artist. I haven't in well over a decade. There's also a canyon of a difference between that, and if I had re-sold their product, at scale, effectively putting the artists out of business. I'd love to explode copyright, but unfortunately, our society has no mechanism for compensating the people that make this valuable work without it, because a whole bunch of tech execs will say "jeez — i'd really like to get paid for their work instead of them."
> There's also a canyon of a difference between that... effectively putting the artists out of business.
There is a direct line between music piracy you did in the past and the status quo of Spotify paying millidollars to artists today. Another POV is, find me musicians who prefer a world with Internet piracy compared to one without.
You're arguing about something I'm not. I completely agree that what I did was wrong, and to boot, I stopped pirating music as soon as online music stores like itunes popped up despite being an impoverished line cook.
That has absolutely no impact, at all, on my fitness to criticize this current wrong.
> Commercial creative workers are vulnerable because there's a billions-of-dollars industry effort to copy their professional output and compete with them selling cheap knock-offs.
I agree there will be winners and losers of some proportion here. But I also think the people that want to pay for art will continue to pay as their motives and values are different. There's plenty of cheap knock-off art, but people still pay premiums for art to support the artist and their work.
As someone else replied to you, it's similar to piracy. The people that pirate were never going to pay in the first place. To tie it back here, the people listening to AI generated <whatever> were never going to pay in the first place - which is why so many podcasts get their money from ads.
> But I also think the people that want to pay for art will continue to pay as their motives and values are different.
The big difference is the type of artist. People selling fine art won't be affected much. The vast majority of artists are commercial artists, the the idea that being a commercial artist is morally or creatively bankrupt— a common sentiment among those who want to imagine that this is all just fine— is nonsense. It's pilfered commercial artwork that makes up the bulk of these tools commercial utility, and the people that made it stand to suffer the most.
I haven't seen that idea (artists being morally bankrupt), like you I'd strongly disagree. I also agree it's a shitty situation that artists invested hundreds of hours of their own time to create something only to be repackaged and sold by some AI tool.
That said, I'd still make the same point that people who value art and the artist will buy from and support the artist. Those that don't value it, won't. But now we're on a larger scale.
>That said, I'd still make the same point that people who value art and the artist will buy from and support the artist.
The chances anyone will come across the artist when their marketplace is flooded with increasingly plausible simulacra become more and more slim as time goes on.
AI is choking off any hope for artists supported by patronage, simply by virtue of discoverability being lost and trust being eroded.
> The people that pirate were never going to pay in the first place.
I think I agree with your larger point, but is this part true? When Spotify provided a much simpler UX to get the goods, people were happy to pay $10/month and Napster et al basically died.
> Tech folks love sentiments like this because it entirely emotionally places the onus on the people getting ripped off by big tech companies for being ripped off.
This, times a million. Add to that the ancient quote from Plato(?) criticizing writing or the other ancient quote complaining about the irresponsibility of the youth, unthinkingly deployed to attempt to delegitimize any kind of critique of nearly anything.
The technology industry seems to be overflowing with so-called "rational" people who mainly seem to use use whatever intelligence they have to rationalize away responsibility for whatever problems their beloved technology has caused. It's a really stupid and obnoxious pattern; and once you see it, it's hard to not see if everywhere and be annoyed.
I think one element of it is naked greed (especially from the entrepreneurs) but I think another big part is a kind of stuntedness and parochialism that's often fueled by overconfidence (because of success in software engineering, forming an identify around "being smart" etc).
It's one of the reasons I left tech altogether after decades. It's like most people in the tech business right now think their totally unique supreme intellectual might gives them enough pan-subject-matter expertise. The further I moved away from development within the business, the more it repelled me.
Nobody is getting "ripped off" by ML models any more than by other humans. When a human wants to launch a high-quality podcast, they survey the market, listen to a lot of other high quality podcasts, and then set to creating their own derivative work.
What ML models are doing is really no different. It's just much, much faster.
Everything humans create is derivative of other works. Speed is the only difference.
The only difference between cracking a 4-bit private key and a 512-bit private key is speed, too. So are private keys of those sizes qualitatively the same thing?
Or is it that, at some nebulous point, a difference in speed between two things impacts the way humans choose to direct their efforts to such a great extent that, for all intents and purposes, the two things are qualitatively different?
> The only difference between cracking a 4-bit private key and a 512-bit private key is speed, too. So are private keys of those sizes qualitatively the same thing?
That's like sayin "The only difference between drinking 1 gallon of water and 100 gallons of water is death." Yes, the quantity of something for a given use-case is bound to give different results.
What the parent comment was commenting is that the actions being taken by these models should not be morally classified as wrong in abundance just as humans following the same process would never be regardless of the output they produced.
I disagree about what the parent is trying to claim.
>What ML models are doing is really no different. It's just much, much faster.
I take this argument to be that (a) what the ML models are doing is fundamentally the same as what humans already do (I agree with this part), (b) that we have no moral problem with humans doing this already (I agree again), and that (c) the fact that AI does it much faster is not sufficient to cause any moral difference (I disagree with this part, and gave a counterexample to show how, in general, a difference in speed can make a moral difference, because that speed difference can have a large impact on how other people decide to behave).
When I was younger, piracy was justified with similar tech folks arguments: "Information wants to be free", "Serves them right for controlling their content in a way that inconveniences me", "If I own a copy, the content is mine to share".
And I used to be right there with them until I realized it was entirely an entirely self-serving way to justify not paying for shit. (I'm not saying it is for everybody, or that everybody's situation mimics mine-- but I was honest enough to admit that's what it was it was for me.) I don't like Adobe's subscription plan, but I was f-ing poor for my younger decades and there's no way in hell I could afford paying a month's rent for photoshop, but $10/mo? I signed up immediately. Also, rather than just using BT whenever I felt like getting an album, I started making deliberate decisions about what albums I wanted and bought them on iTunes when I learned about it in the mid-aughts. Sure the lock-in and DRM suck, but I was happy to pay for the convenience. For indie bands, I still will buy their stuff on Bandcamp even if I can stream it just because they add value to my life and not being legally compelled to pay them isn't the same as not being morally obligated to compensate people whose labor you voluntarily benefit from. I haven't pirated software in decades. If it's not FOSS and I want it, I'll buy it. It's absolutely bananas how many developers make a fat living off of making commercial software but pretend to be radical class warriors when its time to bust out the credit card for anything that isn't physical.
People care about being able to consume information in ways that works for them.
I don't have time to read white papers (nor am I very good at it), but want to know what they consist of. I also want to take my dog for a walk which is hard to do while staring at a screen. This, and other tools like it are useful in achieving that.
> The reason so much writing, podcasting, and music is vulnerable to AI disruption is that quality has already become secondary.
They're vulnerable because people aren't random. Most of what we do can be modeled statistically and translated into patterns and tendencies. Given a sufficient number of parameters, just about anything we do can be digested by an autocompletion program that can then generate an output similar enough to the real thing to fool us.
OK, this is pretty amazing, but is there a "Valley Girl" setting in NotebookLM somewhere? In the sample given in this article, both of the "podcasters" had to add a "like", like every 5 seconds. I couldn't take it:
> this tech is just like leaps and bounds of where it was yesterday like we're watching it go from just spitting out words to like...
"Like" is a filler word I barely notice, along with lower key words like "right" or "uh uh". But the NotebookLM constantly exclaiming "Exactly" and "Precisely" stand out and are driving me a bit loopy. I wonder if you can prompt inject them away.
I would seriously pay, even a subscription fee, to have that ability downloaded into my brain. The first few mentions of "like" don't typically get me, but the more it's used the irritation level grows exponentially...
Strongly suspect it's age-related. I noticed the comment you responded to stated 'along with lower key words like "right" or "uh uh"'.
The turn of phrase "low-key" became popular in the 2010s - I barely, if ever, heard it used before then - so my guess is that this user is in their twenties to early thirties.
Personally, I would love to try this for learning languages.
Some people absorb information far easier when they hear it as part of a conversation. Perhaps it would be possible to use this technique to break down study materials into simple 10-minute chunks that discuss a chapter or a concept at a time.
I am already doing that. What I meant is that there are specific topics, grammar, vocabulary, etc. that I would probably remember better if I also got them in the form of a conversation between two knowledgeable people.
Anyone else thinking that the male voice sounds suspiciously similar to Dax Shepard? I generated one of these podcasts last week and that was the first thing I noticed. I haven't seen any reporting on it.
I was building out something along these lines and the voice was just rubbish (I mean, sounded fine for one short episode but won't on the 20th episode) at the time, so postponed to focus on more near-term goals. But the variation in voices here is quite a bit better, and will improve. You know this is going to be a thing.
There are still some extremely challenging/interesting problems to make it not terrible. This is where we get to invent the future.
They've really nailed the back and fourth of the two speakers!
It would be interesting to know if it's multimodal voice, or just clever prompting and recombining...
I added single voice podcasts to Magpai after seeing how useful this was. Allows for a bit more customisation of the podcast too https://www.youtube.com/watch?v=OEsh9MlbA6s
There are already tons of similar AI-generated content on YouTube. It's only a matter of time before stuff like this becomes the equivalent of the omnipresent SEO spam today.
My first instinct was to not see why one would want to consume such a podcast, a simile instead of either the original or an (AI?) summary of the original.
Then I remembered a partially disabled friend who regularly asks for audio books, because he physically cannot read long form. This, condensed, output would make a lot of ideas accessible to him.
Jesus it's good. I gave it some of my travel blogs, and wow. I mean, there are flaws, particularly in the shallowness of the analysis, but it's at least as good as some time-poor podcast hosts would do.
TBH Im wondering is there anyway to increase the depth or approach by prompting a model for it? Will that be in a future release or hybrid product? ( The audio tech is seemless 100% perfect ) its the quality of the content that needs work now, is there no way to plug this into another LLM ?
510 comments
[ 2.3 ms ] story [ 341 ms ] threadI sent it to my colleagues telling them I "had it produced." I'll reveal the truth tomorrow.
I had a friend who did the same to me, I was sent a message asking my opinion on a tech topic. I spent 30min researching/reading to make sure my reply was accurate and then found out the question was generated by a LLM, and he just wanted to show off how good a LLM was.
It will color every interaction you have with that person...
If a random podcaster says "I've proved that P=NP" I'd say "no you didn't", but if a math professor sends me that same link I'll keep listening to see where this goes. And I've definitely read texts making wild assertions that only at the end were revealed as hit pieces and/or propaganda.
In that case i would listen to all of it aswell, otherwise i can't give honest feedback.
The issue is we would give less attention to these things if it wasn't for the social credit the humans gave the vomit. So we engage in good faith and it turns out it was effectively a prank, and we have no choice but to value requests from those people less now because it was clear they didn't care about our response.
https://www.youtube.com/@TheReportOfTheWeek
Yes, it will generate a middle-of-the-road waffling podcast, but not one with any real depth.
> like thrown into every sentence
I think that's actually part of why it sounds real, because tons of people do actually talk like that.
To me what would make it even better is the ability to throw in random jokes and utilize information about their surroundings and recent events.
I have been using MeloTTS for text-to-speech and I thought that was about the best we could do right now, but apparently I was very wrong. Is there an offline model one can download today that sounds as good as this NotebookLM?
And SoundStorm has more than twice the context window of Bark so dialogs are a tight fit.
When I tried my own text with it, it went completely off the rails... skipping completely over random words, and also switching to different voices in the middle of a sentence. Trying to run the large model also crashed entirely.
Presumably it was trained in noisy data. But it can generate and use a clean voice, they are in there. Most of the Suno default voices are not great either - but a great voice can sound perfectly clear. I haven't done much with Bark lately but on my Twitter there's plenty of clear examples of very realistic voices. Actually here I ran a prompt based on some copy and pasted test 20 times in Bark. I put a couple better results up front, but even in later samples you can find lots of evidence of human-sounding voices. https://sndup.net/bzhz5/
Going off the rails and hallucinating is a hard problem. It can be minimized, but probably would have to solved with simple brute force (check the output with S2T and retry if needed.)
For raw audio you can replace the final decoding step with something like VOCOS or MBD if you want to maximize audio quality, though you don't need do with the best voices.
You can compare it to Google's Illuminate which also generates conversations by summarizing texts but in a much straighter, less fluffy way. It's less shallow but in some ways less compelling:
https://illuminate.google.com/home
Honestly, given the personalization maybe it's a net improvement.
Would they also observe a rocket launch from the grounds of the space center and go "eh, not really impressive" ?
Or maybe they are just defining "impressive" as something totally different to what we're thinking.
Probably calling "impressive" something which adds value and does not suggest eerie bits.
Sam Altman: «They laughed at us... Well they are not laughing now, are they». No, but a different kind of "serious" was raised.
Defending craftsmen and attention to detail is not just about purism or gatekeeping. I appreciate people who care, even in fields I don’t personally care about (yet?). The professor who annoyingly insists on making sure every student “really gets it”, or the woodworker who is adamant about what joints are superior, or the kernel hacker who maintains rigor in face of hundreds of feature requests. The integrity of professionals can make or break institutions.
With AI reducing the effort to create garbage to the point of commoditization, people have a right, and arguably even an obligation, to be concerned. Remember, tech doesn’t follow potential, it follows incentive.
Basically it’s a neat party trick at the moment. I do hope to see it improve however!
> It is interesting that nowadays, practically no one feels that sense of awe any longer - even when computers perform operations that are incredibly more sophisticated than those which sent thrills down spines in the early days. The once-exciting phrase "Giant Electronic Brain" remains only as a sort of "camp" cliché, a ridiculous vestige of the era of Flash Gordon and Buck Rogers. It is a bit sad that we become blasé so quickly.
> There is a related "Theorem" about progress in AI: once some mental function is programmed, people soon cease to consider it as an essential ingredient of "real thinking". The ineluctable core of intelligence is always in that next thing which hasn't yet been programmed. This "Theorem" was first proposed to me by Larry Tesler, so I call it Tesler's Theorem: "Al is whatever hasn't been done yet."
This quote is from the 80s, from GEB by Douglas Hofstadter.
(and btw, I just took a grainy, poorly-lit picture from the book, and could automagically select the text from it, since I couldn't find the quote online. Imagine that tech in the 80s. Hell, it was bad even in the 2000s, with OCR being hit and miss for a long time. Now it "just works".)
Think about how comfortable your life is, and how the 17th century version of yourself would kill to live it. Then think about how you aren't in a perpetual state of ecstasy for being given this life.
People quickly adapt to their current circumstances, take them for granted, and immediately want more.
TBH I think it’s more of a knee jerk reaction from those tired of hearing about AI or who just want to post contrarian opinions (which I totally do sometimes, too).
But I don’t think it’s much of a threat to actual podcasts, which tend to be successful because of the personalities of the hosts and guests, and not because of the information they contain.
Which leads me to hope that the next versions of Notebook will allow more customization of the speakers’ voices, tone, education level, etc.
I wonder if any “blended” podcasts will pop up, where a human host uses a tool like this for an artificial cohost.
I'm not super proud of the Twitter AMA one and if u listen back now i fixed many of the bad cutovers. I doubt i'll repeat it again on current tech.
thank you for listening! feedback and ideas welcome.
> We always start with a clear overview of the topic, you know, setting the stage. You’re never left wondering, “What am I even listening to?” And then from there, it’s all about maintaining a neutral stance, especially when it comes to, let’s say, potentially controversial topics.
Oh yeah, this is exactly why I listen to Oxide's podcast! (This is a joke. They often launch into topics with no explanation or context, and are unabashedly opinionated.)
I do think that this will change in the not too distant future. OpenAI's o1 is a step in the direction we need to go. It will take a lot more test-time compute to produce content that has high quality to match its high production values.
I’d feed it the Singularity paper, but I’m not sure I need that extra boost of anxiety these days…
https://edoras.sdsu.edu/~vinge/misc/singularity.html
In other words: soapbox is presumably some sort of toy car that goes 15mph, and formula 1 goes up above 150mph at least (as you can tell, I’m not a car guy). If you have any actual scientific argument as to why a model that can score 90-100 on a typical IQ test has only 1/10th the symbolic reasoning skills of a human, I’d love to eat my words! Maybe on some special highly iterative, deliberation-based task?
It's also not at all clear to me what "symbolic" could mean in this context. If it means the software has concepts, my response would be that they aren't concepts of a kind that we can clearly recognize or label as such (edit: and that's to say nothing of the fact that the ability to hold concepts/symbols and understand them as concepts/symbols presupposes internal life and awareness).
The best analogy I've heard for these models is this: you take a completely, perfectly naive, ignorant man, who knows nothing, and you place him in a room, sealed off from everything. He has no knowledge of the outside world, of you, or of what your might want from him. But you slip under the door of his cell pieces of paper containing mathematical or linguistic expressions, and he learns or is somehow induced to do something with them and pass them back. When what he does with them pleases you, you reward him. Each time you do this, you reinforce a behavior.
You repeat this process, over and over. As a result, he develops habits. Training continues, and those habits become more and more precisely fitted to your expectations and intentions.
After enough time and enough training, his habits are so well formed that he seems to know what a sonnet is, how to perform derivatives and integrals, and seems to understand (and be able to explain!) concepts like positive and negative, and friend and foe. He can even write you a rap-battle libretto about nineteenth-century English historiography in the style of Thomas Paine imitating Ikkyu.
Fundamentally, though, he doesn't know what any of these tokens mean. He still doesn't know that there's an outside world. He may have ideas that are guiding his behavior, but you have no way of knowing that -- or of knowing whether they bear any resemblance to concepts or ideas you would recognize.
These models deal with tokens similarly. They don't know what a token is or represents -- or we have no reason to think they do. They're just networks of weights, relationships, and tendencies that, from a given seed and given input, generate an output, just like any program, just like your phone keyboard generates predictions about the next word you'll want to type.
Given billions and billions and billions and billions of parameters, why shouldn't such a program score highly on an IQ test or on the LSAT? Once the number of parameters available to the program reaches a certain threshold (edit: and we've programmed a way for it to connect the dots), shouldn't we be able to design it in such a way that it can compute correct answers to questions that seem to require complex, abstract reasoning, regardless of whether it has the capacity to reason? Or shouldn't we be able to give it enough data that it's able to find the relationships that enable it to simulate/generate patterns indistinguishable from real, actual reasoning?
I don't think one needs to be cynical to be unimpressed. I'm unimpressed simply because these models aren't clearly doing anything new in kind. What they're doing seems to be new, and novel, only because of the scale at which they do what they do.
Edit: Moreover, I'm hostile to the economic forces that produced these models, as everybody should be. They're the purest example of what Jaron Lanier has been warning us about -- namely that, when information is free, the wealthiest are going to be the ones who profit from it and dominate, because they'll be the ones able to pay for the technology that can exploit it.
I have no doubt Altman is aware of this. And I have no doubt that he's little better than Elizabeth Holmes, making ethical compromises and cutting legal corners, secure in the smug knowledge that he'll surely pay his mo...
This is very "humans are just hunks of matter! They can't think!".
More specifically, my comment aims to meet the challenge posed by the person I answered:
> I highly recommend letting that fact be a little bit impressive, someday. There’s no way you live through any event that’s more historically significant, other than perhaps an apocalypse or two. [...] If you have any actual scientific argument as to why a model that can score 90-100 on a typical IQ test has only 1/10th the symbolic reasoning skills of a human, I’d love to eat my words.
I have no idea what would constitute a "scientific argument" in this instance, given that the challenge itself is unscientific, but, regardless, the results that so impress this person are, without question, achievable without reasoning, symbolic or otherwise. To say that the model "muses" or "has [...] symbolic reasoning" is to make a wild, arbitrary leap of faith that the data, and workings of these models, do not support.
The models are token-prediction machines. That's it. They differ not in kind but in scale from the software that generates predictions in our cell-phone keyboards. The person I answered can be as impressed as he wants to be by the high quality he thinks he sees in the predictions. That's fine. I'm not. In that respect, we just disagree. But if he's impressed because he thinks the model's predictions must or do betoken reasoning, he's off in la la land -- and so his wide-eyed, bushy-tailed enthusiasm is based on nonsense.
It's no different from believing that your phone keyboard is capable of reasoning, simply because you are delighted that it guesses the 'right' word often enough to please you.
I'll instead say this: if you think these models must be reasoning when they produce outputs that pass reasoning tests, then you should also believe, every time you see a photo of a dog on a computer screen, that a real, actual dog is somewhere inside the device.
You said:
> I'm not saying "they predict tokens; therefore, they can't reason." I'm saying "something that can't reason can predict tokens, so prediction isn't evidence of reasoning."
This is true. Reasoning is evidence of reasoning, and LLMs do pass reasoning tests. Yes, the way they work doesn't imply that they can reason, the fact that they can reason implies that.
You also said:
> These models deal with tokens similarly. They don't know what a token is or represents -- or we have no reason to think they do.
I have no reason to think that other people know what concepts are or what they represent, just that they can convincingly output a stream of words when asked to explain a context.
The argument bugs me because you can replace "they predict the next token" with "they are collections of neurons producing output voltages in response to input voltages" and you'll have the exact same argument about humans.
Here I think it's helpful to distinguish between what something is and how it's known. When we see something that resembles reasoning, we very reasonably deduce that reasoning has taken place. But 'it looks like reasoning' is not equivalent to 'it is reasoning.'
To approach the same idea from a different direction:
> I have no reason to think that other people know what concepts are or what they represent, just that they can convincingly output a stream of words when asked to explain a context.
You absolutely do have reason to think this. You're the reason. You're the best available evidence, because you have an internal life, have concepts and ideas, have intentions, and perform acts of reasoning that you experience as acts of reasoning -- and all of that takes place inside a body that, you have every reason to think, works the same way and produces the experiences the same way in other people.
So, sure, it's true that you can't prove that other people have internal lives and reason the way you do. (And you're special, after all, because you're at the center of the universe -- just like me!) But you have good reason to think they do -- and to think they do it the way you do it and experience it the way you experience it.
In the case of these models, we have no such reason/evidence. In fact, we have good reason for thinking that something other than reasoning as we think of it takes place. We have good reason, that is, to think they work just like any other program. We don't think winzip, Windows calculator, a Quake bot, or a piece of malware performs acts of reasoning. And the fact that these models appear to be reasoning tells us something about the people observing them, not about the programs themselves. These models appear to be reasoning only because the output of the model is similar enough to 'the real thing' for us to have trouble saying with certainty that they aren't the real thing. They're simulations whose fidelity is high enough to create a feeling in us -- and to pass some tests. (In that sense, they're most similar to special effects.) (Edit: and that's not to say feelings are wrong, invalid, or incorrect. They're one of the key ways we experience the things we understand.)
Is reasoning taking place in these models? Sure, it's possible. Is there an awareness or being of some kind that does the reasoning? Sure, that's possible, too. We're matter that thinks. Why couldn't a program in a computer be matter that thinks? There's a great novel by Greg Egan, Permutation City, that deals partly with this: in one section, our distant descendants pass to another universe, where matter superficially appears to be random, disorganized, and low in enthalpy. When that random activity and apparent lack of life and complexity are analyzed in the right way, though, interference patterns are revealed, and these contain something that looks like a rich vista bursting with directed, deliberate activity and life. It contains patterns that, for all the world, look and act like the universe we know -- with things that are living and things that are not, with ecosystems, predators, prey, communities, reproduction, etc. These patterns aren't in, and aren't expressed in, the matter itself. They 'exist' only in the interference patterns that ripple through it.
That's 100% plausible, too. Why couldn't an interference pattern amount to a living thing, an organism, or an ecosystem? The boundary we draw between hard, physical stuff and those patterns is arbitrary. Material stuff is just another pattern.
My point isn't that reasoning doesn't take place in these models or can't. It's, first, that you and I do something we call reasoning, and the best available information tells us these models aren't doing that. Second, if they are doing something we ...
And this is equivalent to saying there's a dog in my computer when I open a photo of a dog. It treats the simulation, the data, the program -- whatever you want to call it -- as if it were the thing itself.
https://en.wikipedia.org/wiki/SHRDLU
Because, whenever you give it a reasoning test, it also seems to do fine.
That is what I meant in my other post, I don't really think that "it seems to do fine" is enough evidence for the extraordinary claim that it can reason.
Therefore I think it's reasonable to say they cannot, until someone actually comes up with that evidence.
They are essentially next token predictors after first training, but then instruct models are fine tuned on reasoning and Q/A scenarios, afaik early research has determined that this isn't just pure parroting, that it does actually result in some logic in there as well.
People also have to remember the training for these is super shallow at the moment, when compared with what humans go through in our lifespans as well as our millions of years of evolution (as humans).
I'm saying, rather, that the models do what they're taught to do, and what they're taught to do are computations that give us a result that looks like reasoning, just the way I could use 3ds max as a teenager to generate on my computer screen an output that looked like a cube. There was never an actual cube in my computer when I did that. To say that the model is reasoning because what it does resembles reasoning is no different from saying there was an actual cube somewhere in my computer every time I rendered one.
Say you need to read those instructions, but it’s also really nice out and you want to go for a jog: two birds, one stone.
As a podcast listener, I lose interest if I can tell the audio is AI-generated...
If you want you can do a test with people that haven't heard about the tech. Have it generate something you know they'll enjoy, maybe 2-3 min long, and have them listen to it without knowing it's AI generated. Ask them about the subject, and see if anyone mentions anything about being fake. You'll be surprised.
You would however be able to tell that it was extremely obnoxious and bland, and without the novelty of the technical trick, you would not be listening to it.
I've never naturally come across a podcast that's AI generated to have this reaction.
I imagine it'll be even harder to know if regular pod casters feed the AI a few episode they've made, to make it learn how to talk like they do. Like, would you really know if your true crime pod casters skipped a week if the AI sounded like them? I guess I don't really fall into the category for your question as I'm not a pod cast listener.
This is in-line with all art, music, and video created by LMM at the moment. They are imitating a structure and affect, the quality of the content is largely irrelevant.
I think the interesting thing is that most people don't really care, and AI is not to blame for that.
Most books published today have the affect of a book, but the author doesn't really have anything to say. Publishing a book is not about communicating ideas, but a means to something else. It's not meant to stand on its own.
The reason so much writing, podcasting, and music is vulnerable to AI disruption is that quality has already become secondary.
I’ve heard at least one ad from dozens (probably a hundred) podcast episodes that I didn’t finish.
In terms of such music and films (whether created by human or AI) sometimes it's just because we are social creatures and need shared experiences to talk with others about.
In reality, between work, sleep and family, I rarely have anything resembling that kind of time and mental energy reserve available.
But, what I can afford is to listen to podcasts while doing other things. Doing that gives me enough of an overview to keep up with a general topic and find new topics that might be worth investing into deeper.
Wouldn’t it be great if someone made a podcast channel specifically for “Papers corysama wants to hear about at this moment”? I think so. Apparently, so do a lot of other people. But, they don’t want to list to my specific channel.
Reading a book is a time investment so I want it to convey the thoughts of another human being, otherwise it would feel like wasting my time. Listening to music, on the other hand, often is something that I do while I exercise, to keep a brisk pace and not get bored. As long as it sounds good, fits the genres and styles I like and is upbeat enough for exercising, I wouldn't have much of a problem with AI music - maybe it would even be a plus, since there are some specific music genres where I have already listened to pretty much everything there is (and no more is being made), and it would be great to have more.
I don't listen to podcasts, but I suppose in that case it depends on how you do so: devoting your full time and attention like a book, or as a background while you do something else like exercise music? As far as I know, many listeners are in the latter case, so I don't see why they wouldn't listen to AI podcasts.
I see the same with potteries. A factory made pot cannot have more value than a hand made pot with the signature of a human. This touches the very fabric of society. Hard to explain.
Too much of it -> No, there are entire musical genres (e.g. italodance or big beat) where I have already listened to pretty much everything available, and they are not expanding anymore because they are not fashionable. It would be nice to have more songs and be surprised.
Isn't novel in any way -> This is not how it works, there are studies showing that AI can be creative. Or at the very least (since the definition of creativity can be controversial) produce output that is indistinguishable from novel, creative output, which is enough for the purpose discussed here.
Until we get better versions of o1 that can generate insights over days and then communicate them in book form the loss of interactivity and personalisation makes LLM books pointless.
I do suspect that interactive media is just strictly better in theory. But maybe there will be a period of time where bespoke AI-generated books make sense.
Why not just seek out the original works that the AI stole from?
I know that you likely intended to imply that you can subsitute the aformentioned AI music with an existing piece of music of the same genre, but that is not a view shared by all. Sometimes the generated music scratches such a specific and personal itch, that it cannot be replicated by something in the same genre.
A better counterargument to your original comment would be "It is not an exclusive situation. I can listen to and support both generated music and handcrafted music at the same time. They both contain music tracks that I like."
To be more specific about the second sentence, if there are any readers in doubt:
> The generated music wouldn't exist without the foundation of stolen music made by people.
The word "stolen" is a value judgement that is not shared by all. It is a word meant to invoke an emotional response in the reader. For example, Stallman has argued that the data could not have been stolen, or else it would not be there anymore. So, removing this word gives you:
> The generated music wouldn't exist without the foundation of existing music made by people.
Which is a true fact that has never been in debate.
However, this is not relevant to the main point that not all generated music has a suitable handcrafted substitute, and that there is no actual need to choose exclusively to listen to generated or human crafted music. Furthermore, the conversation has turned uncivil (the first sentence). Therefore, goodbye.
A randomly selected NotebookLM podcast is probably not substantial enough on its own. But with human curation, a carefully prompted and cherry-picked NotebookLM podcast could be pretty good.
Or without curation, I would use this on a long drive where audio was the only option to get a quick survey of a bunch of material.
So where does AI regurgitated slop fit into my life?
For example, Perun. This guy delivers an hourlong presentation on (mostly) the Ukraine-Russia war and its pure quality. Insights, humour, excellent delivery, from what seems to be a military-focused economist/analyst/consultant. We're a while away from some bot taking this kind of thing over.
https://www.youtube.com/@PerunAU
Or hardcore history. The robots will get there, but it's going to take a while.
https://www.dancarlin.com/hardcore-history-series/
Thankfully there's plenty out there I am a fan of!
Carlin on the other hand, despite the digressions and rambling, manages to keep you engaged and really feel the events.
I tried a few episodes. I really tried. I couldn’t do it. It reminded me of my uncle would tell a 5 min story in half an hour.
The dramatic filler, breathless story telling was too much for me. If anything it would put me to sleep.
I’ve found a few history podcasts that I think go into a lot more depth and I learn a lot more from.
If you just need voice discussing some topic because that has utility and you can't afford a pair of podcasters (damn, check your couch cushions) then having a mid podcast is better than having no podcast. But if you need expert Insight because expert Insight is your product and you happen to deliver it through a podcast then you need an expert.
If I were a small software shop and I wanted something like a weekly update describing this week's updates for my customers and I have a dozen developers and none of us are particularly vocally charismatic putting a weekly update generated from commits, completed tickets, and developer notes might be useful. The audience would be very targeted and the podcast wouldn't be my main product, but there's no way I'd be able to afford expert level podcasters for such a position.
I would argue Perun is a world class defense Logistics expert or at least expert enough, passionate enough, and charismatic enough to present as such. Just like the guys who do Knowledge Fight, are world class experts on debunking Alex Jones, and Jack Rhysider is an expert and Fanboy of computer security so Darknet Diaries excels, and so on...
These aren't for making products, they can't compete with the experts in the attention economy. But they can fill gaps and if you need audio delivery of something about your product this might be really good.
Edit - but as you said the robots will catch up, I just don't know if they'll catch up with this batch of algorithms or if it'll be the next round.
I've seen people manage to wrangle tools like Midjourney to get results that surpass extra medium. And most human artists barely manage to reach medium quality too.
The real danger of AI is that, as a society, we need a lot of people who will never be anything but mediocre still going for it, so we can end up with a few who do manage to reach excellence. If AI causes people to just give up even trying and just hit generate on a podcast or image generator, than that is going to be a big problem in the long run. Or not, and we just end up being stuck in world that is even more mediocre than it is now.
Do we though? That seems bleak.
I guess if AIs become excellent at everything, and the gains are shared, and the human race is liberated into a post-scarcity future of gay space communism, then it's fine. But that's not where it's looked like we're heading so far, though - at least in creative fields. I'd include - perhaps not quite yet, but it's close - development in that category. How many on this board started out writing mid-level CRUD apps for a mid-level living? If that path is closed to future devs, how does anyone level up?
I think one of the major reasons this is the case is because people think it's just not possible; that the way we've done things is the only possible way we can continue to do things. I hope that changes, because I do believe AI will continue to improve and displace jobs.
That may be your position as well - indeed, I think your point about "people think[ing] it's not possible" is directly relevant - but I wanted to make that more explicit than I did in my original comment.
It'd be like the ancient Romans speculating that cars will make us less fit and therefore cities will be less impressive because we can't lift as much. That isn't at all how it played out, we just build cities with machines too and need a lot less workers in construction.
There are a lot of ways if it did reach intellectual excellence that we could argue that it would make Humanity more mediocre, I'm not sure I buy such arguments but there are lots of them and I can't say they're all categorically wrong.
Isn’t this exactly how it played out?
We move orders of magnitude more cargo and material than them because fitness isn't the limiting factor on how much work gets done. They didn't understand that having humans doing all that labour is a mistake and the correct approach is to use machines.
[0] Given the floods I saw recently, I'm not even sure this is even true.
If someone wants to click generate on a podcast button or image generator it seems unlikely to me that was a person who would have been sufficiently motivated to make an excellent podcast or image. On the flip side, consider if the person who wants to click the podcast or image button wants to go on to do script writing, game development, Structural Engineering, anything else but they need a podcast or image. Having such a button frees up their time.
Of course this is all just rhetorical and occasionally someone is pressed into a field where they excel and become a field leader. I would argue that is far less common than someone succeeding and I think they want to do, but I can't present evidence that's very strong for this.
This will be the dynamic of generated art as it improves; the ease of use will benefit creators at the fringe.
I bet we see a successful Harry Potter fanfic fully generated before we see a AAA Avengers movie or similar. (Also, extrapolating, RIP copyright.)
Or to put another way, I've heard much better ideas on a podcast made by undergrad CS students than on Lex Fridman.
https://www.pewresearch.org/journalism/fact-sheet/cable-news...
(I suppose you could also quibble with "rapidly".)
We all seek different kinds of quality; I don't find Peruns videos to have any quality except volume. He reads bullet points he has prepared, and makes predictable dad jokes in monotone, re-uses and reruns the same points, icons, slides, etc. Just personally, I find it really samey and some of the reporting has been delayed so much it's entirely detached from the ground by the time he releases. It's a format that allows converting dense information and theory to hour long videos, without examples or intrigue.
Personally, I prefer watching analysis/sitrep updates with the geolocations/clips from the front/strategic analysis which uses more of a presentation (e.g. using icons well and sparingly). Going through several clips from the front and reasoning about offensives, reasons, and locations is seems equally difficult to replicate as Peruns videos, which rely on information density.
I do however love Hardcore history - he adds emotion and intrigue!
I agree with your overall hope for quality and different approaches still remaining stand out from AI generated alternatives.
Give it a year or three, up to the minute AI generated sitrep pulling in related media clips and adding commentary…not that hard to imagine.
But why? Isn’t there enough content generated by humans? As a tool of research AI is great in helping people do whatever they do but having that automated away generating content by itself is next to trash in my book, pure waste. Just like unsolicited pamphlets thrown at your door you pick up in the morning to throw in the bin. Pure waste.
https://m.youtube.com/@militaryandhistory
https://m.youtube.com/@suchomimus9921
https://m.youtube.com/@WardCarroll
Fav is probably Suchomimus right now. Updated faster, shorter reports. I feel like I get the info sooner after it happens.
I like Hardcore history very much, but I think it would be far worse in a video form.
I listen to Perun at the gym every week, audio only.
AFAIK, it's only a paid feature to play video in the background.
The presentation is a matter of taste (I like it better than you do), but the content is very informative and insightful.
Its not really about what is happening at the frontline right now. Its not its aim. Its for people who want dense information and analysis. The state of the Ukrainian and Russian economies (subjects of recent Perun videos) does not change daily or weekly.
And then faster/easier/cheaper access to the LM 'uninspired but possibly useful' content, whatever that might look like.
I suspect LLMs are not sophisticated enough as a paradigm to get there.
It's an article of faith -- we don't KNOW that they're going to get there. They're going to get better, almost certainly, but how much? How much gas is left in the tank for this technique?
Honestly, I think the fact that every new "groundbreaking" news release about LLMs has come alongside a swath of discussion about how it doesn't actually live up to the hype, that it achieves a solid "mid" and stops there, I think this means it's more likely that the robots AREN'T going to get there some day. (Well, not unless there's another breakthrough AI technique.)
Either way, I still think it's interesting that there's this article of faith a lot of us have "we're not there now, but we'll get there soon" that we don't really address, and it really colors the discussion a certain way.
If you think about how an LLM works, it's effectively going "given a certain input, what is the statistically average output that I should provide, given my training corpus".
The thing is, humans are remarkably shit at understanding just how exception someone needs to be to be genuinely creative in a way that most humans would consider "artistic"... You're talking 1/1000 people AT best.
This creates a kind of devils bargain for LLMs where you have to start trading training set size for training set quality, because there's a remarkably small amount of genuinely GREAT quality content to feed this things.
I DO believe that the current field of LLM/LXM's will get much better at a lot of stuff, and my god anyone below the top 10-15% of their particular field is going to be in a LOT of trouble, but unless you can train models SOLELY on the input of exceptionally high performing people (which I fundamentally believe there is simply not enough content in existence to do), the models almost by definition will not be able to outperform those high performing people.
Will they be able to do the intellectual work of the average person? Yeah absolutely. Will they be able to do it probably 100/1000x faster than any human (no matter how exceptional)?... Yeah probably... But I don't believe they'll be able to do it better than the truly exceptional people.
A decent LLM can just keep going. Time and stamina are effectively unlimited, and an LLM can just keep rolling its 100 dice until they all come up sixes.
Or an author can just input their ideas and have an LLM do the boring bit of actually putting the words on the paper.
What's that saying? "Nobody ever went broke overestimating the poor taste of the average person"
Not replacement, but ecosystem collapse.
I am not saying it's not a sad state of affairs. I am just saying we have been there for a while and the floor might be raised, a bit at least.
I’m not sure we really want that, but I am pretty sure we’ll try for it.
There's still much to be done re: reorganizing how we behave such that we can reap the benefits of such a competent helper, but I don't think we'll be handing the reigns over any time soon.
- "Given this thing I saw a computer program do, clearly we'll have intelligent AI real soon now."
- "If we generate sufficiently smart AI then clearly all the jobs will go away because the AI will just do them all for us"
- "We'll clearly be able to do the AI thing using a reasonable amount of electricity"
None of these ideas are "clear", and they're all based on some "futurist faith" crap. Let's say Microsoft does succeed (likely at collosal cost in compute) in creating some humanlike AI. How will they put it to work? What incentives could you offer such a creature? What will it want in exchange for labor? What will it enjoy? What will it dislike? But we're not there yet, first show me the intelligent AI then we can discuss the rest.
What's really disturbing about this is hype is precisely that this technology is so computationally intensive. So of course the computer people are going to hype it--they're pick and shovel salespeople supplying (yet another) gold rush.
An American Quakening
https://youtu.be/wGpdxsgreOE?si=r7ef1vBOjIvqD_PQ
I think this is just partly an inevitable consequence of going from "content scarcity" to our new normal of "content obesity" over the past 20 years or so. In this new era of an overwhelming amount of content, it's just natural to compare it all against each other, e.g. to essentially "optimize" it to the "best" form, but in doing that we've fallen into a homogeneity, and the resulting lack of variation is an actual lowering of quality in and of itself.
2 examples to explain what I mean:
1. I find that nearly all interior design (at least within broad styles) looks basically the same to me now. It's all got that "minimalist, muted tones but with a touch of organic coziness and one or two pops of color" look to it. Honestly, I don't know how interior designers even exist today, when it's trivial to go to Houzz or any of a million websites and say "yes, like this". A while back I was complaining online somewhere that I thought all interior design looked similar where in the past there was much more interesting variation, and somebody insightful replied that it's not really that interior design is now just the same, it's that it's really just converged. People can easily see and compare a million designs against each other, so there is much less of a chance for that green shag carpet to even get a moment in the sun.
2. I was recently on vacation and decided I wanted to read a "classic" book, so I read Hemingway's The Sun Also Rises (I'm not sure why I never had to read that in high school). Nearly throughout the entire book I couldn't help but thinking "Is there any time this book stops sucking?" I hated the entire thing - it was like being forced to watch someone's vacation photos for twelve hours straight, and I kept wondering why there never seemed to be any attempt to actually make me give a shit about any of the characters in the book, as nearly every one of them I found insufferable and wondered how they each had about 3 or 4 livers to spare. But I do understand that Hemingway's writing style was unique and original at the time, and that he was doing something new and interesting that influenced American literature for a long time. But these days, given the flood of content, it feels like most attempts at doing something "new and interesting" are not only forced, but nearly impossible given that there are a million other people also trying to do new and interesting things that now have the means to disseminate them. I don't think a book like The Sun Also Rises, where I believe the main impact was the style of writing/dialogue vs the actual story, could ever break through today.
I guess my point with this long post is that I think the "loss of quality" in content that many of us sense is just a direct result of there being so much content that we see variations from the "ideal" as worse, where in the past we may have found them interesting.
Still, at the same time, I couldn't help but feeling a little bit sad/resigned at the existence of the article you linked. Here I thought I had an idea that was not exactly unique but that I felt would be good to share. And yet then here is an example that expresses this idea a million times better than I ever could (I love "The Age of Average" headline), with great researched examples and tons of helpful visuals. It's hard to not feel a bit like Butters in that "Simpsons did it!" episode of South Park...
A lawyer friend of mine also suggested giving it the Spanish civil code, a long, arid legal text. The podcast of course didn't cover the whole text in 10 minutes, which would be impossible, but they selected some interesting tidbits and actually had me hooked until the end and made me learn a few things about it, which is no small merit. And my friend was quite impressed and didn't complain about correctness.
Lately, when I just want to get the gist of a long article or research paper, I run it through NotebookLM and listen to the podcast while I’m exercising.
My only complaint is that the chatty podcasty gab gets tiring after a while. I wish it were possible to dial that down.
Reading, or listening to podcast, these days is more akin to a meditation - many people do it to reenforce an identity rather than to expand on themselves.
And I do think that is reasonable as, for many people, there are few other structures that can keep them in check with themselves.
In my company, HR now uses AI to do training videos. It's hilariously funny, because it looks like a satire on training videos (well, granted, it's funny for a minute or two, then it shifts to annoying).
If anything, listening to that reminded me of why I stopped listening to podcasts in the first place - every 5 second snippet of something interesting ends up suffocated by 5 minutes of filler and dead air.
Think back to the mid-1980s and the first time everyone got their hands on a Casio or Yamaha keyboard with auto-accompaniment.
It was a huge amount of fun to play with, just pressing a few buttons, playing a few notes and feeling like you were producing a "real" pop song. Meanwhile, any actual musicians were to be found crying in the corner of the room, not because a new tool had come along which threatened their position, but because non-musicians apparently didn't understand (at least immediately) the difference between these superficial, low-effort machine-generated sounds and actual music.
It isn't so black and white.
https://youtube.com/shorts/Wn0NtSNeQEQ
[1]: https://en.wikipedia.org/wiki/Sleng_Teng
People keep forming these analogies/explanations with the inherent premise that what we have now is what AI is going to be - "It's actually kind of shitty so don't fret, not much will change".
AI music creation has improved more in the last 5 years than keyboard accompaniment improved in the previous 40 years. It would be very brazen to bet that the tech 5 years from now is hardly any better. Especially when scaling transformers has consistently improved outputs. Double especially when the entire tech industry is throwing the house at scaling it.
The reason why people like music is because another person wrote and performed it. We like watching other people.
Give us an infinite playlist of elevator music and it just becomes oatmeal.
Popular music has already been synthetic and souless for decades now. People will listen to what sounds good to them, and we already know the bar is very low, and that the hard truth is that it is all subjective anyway.
https://www.nydailynews.com/2016/05/29/thousands-of-new-york...
You just proved my point for me.
https://legacy.iftf.org/future-now/article-detail/making-mik...
We’ve had software accompaniment for a long time. Elevator music. The same 4 chords arranged in similar ways for decades. Hasn’t destroyed music. Neither will AI.
At some point people are going to want to know who’s on the other side making the music.
Unless your argument is that nobody values artists… which is I guess one of the primary conceits of GenAI enthusiasts today.
But the content here has been fed into it deliberately.
That's definitely going to be an improvement. Not.
I think that has always been the case, we just tend to compare today’s average stuff with the best stuff from earlier days.
For example, most furniture pictures from the 60s and 70s are from upper middle class homes. If we listen music, we listen to Queen and not some local band from Alabama (not that I’m against such bands at all; they can make great music too).
I agree with this of course, because generally nobody remembers the bad stuff unless it was the worst. I beg to differ with music, though, because there's an opposing effect: we tend to be left with the most marketed music, which was usually a cheap knockoff of something interesting going on at the time. The shitty commercial knockoff becomes the "classic" while the people they were ripping off don't even get a wikipedia page.
If you ask most people, they are by definition more likely to connect with broadly disseminated cheap knock offs than they are with whatever 'legit' inventive underground creator, simply because they've heard the former and not the latter.
Just a mental exercise: If you ask 1000 people if they prefer Knock Off or Original, and 900 say Knock Off, which one was better? If the answer is still Original, by what metric do we measure quality?
This has been the case as far back as I began reading books which is about 30 years.
Podcasts are only somewhat about things. The most important part is that they're by people, and the people is what draws people in. These ai podcasts are not by people, and when you listen to more than one you start to see the patterns and void where a personality is.
... but it's also pointless. And it's likely different episodes on different topics will tend to sound very much alike; it's already the case here, I'm sure I heard another example where the two voices were the same.
In less than a year we all have learned to recognize AI images with pretty good accuracy; text is more difficult, but podcasting seems easy in comparison.
It ends up sounding like a smarmy Sunday-morning talk show conversation, with over-exaggerated affect and no content.
So far I've just fed it technical papers, which may be part of the problem, but what I got back was, "Gosh, imagine if a recommender system really understood us? Wow, that would be fantastic, wouldn't it?"
https://www.youtube.com/watch?v=ssDdqq_9TzI&t=34s [April Ludgate meets Tynnifer, Parks and Rec]
no one has gotten feedback from "most people" .. this is raw hyperbole
Commercial creative workers are vulnerable because there's a billions-of-dollars industry effort to copy their professional output and compete with them selling cheap knock-offs.
I see this sort of convenient resignation all the time in the tech crowd... "creative workers only can blame themselves for tech companies taking their income because their art just isn't any good anymore!"
The poor quality "content" that's been proliferating recently has been created, largely, using the very tools that AI has built, or their immediate precursors. AI, for all its benefits, has only made that worse.
If you're saying, in good faith, that most of the infomercials, televangelist programs, talk radio, celebrity autobiographies, self-help books, scandalous expose books, and health/exercise fad books etc etc etc that came out 50 years ago were made for no reason beyond advancing human knowledge, you're either too young to remember any media from before our current era and haven't looked beyond survivorship bias.
Tech folks love sentiments like this because it entirely emotionally places the onus on the people getting ripped off by big tech companies for being ripped off. If their work was that awful, companies wouldn't be clamoring to vacuum it up into their models to make more of it. Nearly all of the salable output from these models exists solely because it took a creative product someone made with the intention of selling it and it's using it to sell a simulacra.
It's using nostalgia to deflect guilt for harpooning the livelihood of many people because it's just more convenient and profitable to empower mediocre "content creators" they use to justify doing it.
Is that not the work of commercial creative workers? Did it not exist pre AI? There's an argument to scale, certainly, but the idea that "things were better in the past before these <<new technologies>> came out" is generally a suspect argument.
To your broader point - new tools for creating creative work come out all the time. Did we suffer greatly at the loss of image compositors when Photoshop arrived? On the flip side, did digital art gut painting and sculpture? Isn't this just another tool for creative expression?
Art is a way of seeing, not a way of creating. I don't think the technology is taking that away.
The fact that all of that stuff was crap is central to my point. You might just need to give it another read.
> Art is a way of seeing, not a way of creating. I don't think the technology is taking that away.
I'm really sick and tired of the tech industry's bumper-sticker-level-reductive pseudo-philosophical generalizations about "what art is," what it means to be an artist, the acceptable ways to be an artist, and all of that. Art is a whole fucking lot of things, and chief among them in this context is a class of professions. Glib decrees based on a razor-thin slice of one of the broadest topics in the human experience that conveniently exclude or dismiss the stakes of those with the loudest criticism and the most to lose is obviously self-serving. If you're going to take the libertarian "well that's the market for ya" stance," at least be honest about it. If you're going to try to carefully define the entire universe of ideas and practices that comprise art to conveniently exclude the concerns of the people getting screwed over because you think the optics are better or you feel less icky about it, well you better expect to get some really pissed off responses from them.
You can disagree! Folks who are impacted have every right to be pissed, organize, take action. All of these creative endeavors existed _post technology updates_ though - that's my entire point. The need for art doesn't disappear - it changes. Standing athwart the change is a choice, but I'm not sure it is an effective position.
>> Art is a way of seeing, not a way of creating.
> There's no glib decree
This is a glib decree and it completely ignores most of what art actually is in our world, rather than the quaint little box that most people in the NN business try to stuff it into. Your patronizing tone doesn't lend any authority or add depth to your initial analysis, which you essentially just restated using more words. The "art vs craft" dichotomy doesn't even approach the depth and complexity of the interplay of art and commerce in the worlds like video game development, music, cinema and television, and writing... hell even advertising. Like most tech dudes that assume their incredible mental might gives them some kind of pan-topic expertise allowing them to casually dismiss subject matter experts in other fields based on a few a priori thought exercises, you simply don't know how much more you need to learn to make informed decisions about this topic.
There is a direct line between music piracy you did in the past and the status quo of Spotify paying millidollars to artists today. Another POV is, find me musicians who prefer a world with Internet piracy compared to one without.
That has absolutely no impact, at all, on my fitness to criticize this current wrong.
I agree there will be winners and losers of some proportion here. But I also think the people that want to pay for art will continue to pay as their motives and values are different. There's plenty of cheap knock-off art, but people still pay premiums for art to support the artist and their work.
As someone else replied to you, it's similar to piracy. The people that pirate were never going to pay in the first place. To tie it back here, the people listening to AI generated <whatever> were never going to pay in the first place - which is why so many podcasts get their money from ads.
The big difference is the type of artist. People selling fine art won't be affected much. The vast majority of artists are commercial artists, the the idea that being a commercial artist is morally or creatively bankrupt— a common sentiment among those who want to imagine that this is all just fine— is nonsense. It's pilfered commercial artwork that makes up the bulk of these tools commercial utility, and the people that made it stand to suffer the most.
That said, I'd still make the same point that people who value art and the artist will buy from and support the artist. Those that don't value it, won't. But now we're on a larger scale.
The chances anyone will come across the artist when their marketplace is flooded with increasingly plausible simulacra become more and more slim as time goes on.
AI is choking off any hope for artists supported by patronage, simply by virtue of discoverability being lost and trust being eroded.
>But now we're on a larger scale.
It's simply a bad problem, made worse!
I think I agree with your larger point, but is this part true? When Spotify provided a much simpler UX to get the goods, people were happy to pay $10/month and Napster et al basically died.
This, times a million. Add to that the ancient quote from Plato(?) criticizing writing or the other ancient quote complaining about the irresponsibility of the youth, unthinkingly deployed to attempt to delegitimize any kind of critique of nearly anything.
The technology industry seems to be overflowing with so-called "rational" people who mainly seem to use use whatever intelligence they have to rationalize away responsibility for whatever problems their beloved technology has caused. It's a really stupid and obnoxious pattern; and once you see it, it's hard to not see if everywhere and be annoyed.
I think one element of it is naked greed (especially from the entrepreneurs) but I think another big part is a kind of stuntedness and parochialism that's often fueled by overconfidence (because of success in software engineering, forming an identify around "being smart" etc).
Nobody is getting "ripped off" by ML models any more than by other humans. When a human wants to launch a high-quality podcast, they survey the market, listen to a lot of other high quality podcasts, and then set to creating their own derivative work.
What ML models are doing is really no different. It's just much, much faster.
Everything humans create is derivative of other works. Speed is the only difference.
Or is it that, at some nebulous point, a difference in speed between two things impacts the way humans choose to direct their efforts to such a great extent that, for all intents and purposes, the two things are qualitatively different?
That's like sayin "The only difference between drinking 1 gallon of water and 100 gallons of water is death." Yes, the quantity of something for a given use-case is bound to give different results.
What the parent comment was commenting is that the actions being taken by these models should not be morally classified as wrong in abundance just as humans following the same process would never be regardless of the output they produced.
>What ML models are doing is really no different. It's just much, much faster.
I take this argument to be that (a) what the ML models are doing is fundamentally the same as what humans already do (I agree with this part), (b) that we have no moral problem with humans doing this already (I agree again), and that (c) the fact that AI does it much faster is not sufficient to cause any moral difference (I disagree with this part, and gave a counterexample to show how, in general, a difference in speed can make a moral difference, because that speed difference can have a large impact on how other people decide to behave).
I was thinking this kind of thing is the perfect way to generate sports commentary.
I don't have time to read white papers (nor am I very good at it), but want to know what they consist of. I also want to take my dog for a walk which is hard to do while staring at a screen. This, and other tools like it are useful in achieving that.
They're vulnerable because people aren't random. Most of what we do can be modeled statistically and translated into patterns and tendencies. Given a sufficient number of parameters, just about anything we do can be digested by an autocompletion program that can then generate an output similar enough to the real thing to fool us.
I sent the podcast audio to friend, and English is not their first language. Without telling them it was AI generated.
They found it entertaining-worthy enough to listen to the end.
Sure it needs more human unpredictably and some added goofiness. Maybe some interruptions because humans do that too. But it's already not-bad.
> this tech is just like leaps and bounds of where it was yesterday like we're watching it go from just spitting out words to like...
In its current version, this causes so much cultural dissonance that it’s very difficult for me to listen.
At least to me the “hosts” appear to actively signal lack of competence in the field they are talking about.
Given that they are generated that is off course nonsense.
The turn of phrase "low-key" became popular in the 2010s - I barely, if ever, heard it used before then - so my guess is that this user is in their twenties to early thirties.
Some people absorb information far easier when they hear it as part of a conversation. Perhaps it would be possible to use this technique to break down study materials into simple 10-minute chunks that discuss a chapter or a concept at a time.
There are still some extremely challenging/interesting problems to make it not terrible. This is where we get to invent the future.
It would be interesting to know if it's multimodal voice, or just clever prompting and recombining...
I added single voice podcasts to Magpai after seeing how useful this was. Allows for a bit more customisation of the podcast too https://www.youtube.com/watch?v=OEsh9MlbA6s
I've got a daily podcast of hackernews being generated here too: https://www.magpai.app/share/n7R91q
>do you have tools to detect if audio is generated by notebooklm?
>we’re seeing a rise in fake, single-episode podcasts submitted to http://listennotes.com using it.
That seems the direction we're headed in, and some people say the zipbombers can't come soon enough.