> couple of companies are already executing in these areas and have amassed loyal followings
What companies are there in the interactive literotica space?
I was curious about building something like this myself (before people dismiss this as a dumb idea, they should check out what the best selling books on Kindle are).
Fiction.live is going to be huge in this area, they can leverage their existing content and writers to get incremental results.
Since the technology is in its early days the writers can manually improve the generated results to get it to be useful quality, and this will tide the whole system over until perfect completely-AI results are here, guessing sometime towards the end of the year.
Well, the simplest of 'literotica' is just an uncensored, fiction-writing friendly LLM (Fine tuned on fiction/RP, rather than truthful and friendly responses).
NovelAI has been around for a long time, after AI dungeon died in flames. They sell exactly that. They also did the first powerful fine-tune of SD. Now they've got a H100 cluster, and are committed to training a GPT3.5 equivalent.
These companies have difficulties getting investor funding, the good side is however, AI products are so useful, users have an extremely high intention to pay. So these no-investor companies can pure bootstrap themselves with high profitability and high growth.
Based purely on interactions with ChatGPT, interactive fiction isn't really there yet. ChatGPT is not able to tell a cohesive interactive story, because it doesn't understand what's going on. Just for one example, it forgets who is where, and has characters who are not present ("wait in the car") participate in dialogue.
I'm not sure if this is something that can be iterated on, or if it's a fundamental characteristic of large language models.
this is being addressed as a priority, it's a lack of long term memory, you can improve results by repeating key facts during the conversation, stating the same request in different ways and by asking it to summarise unique elements of the conversation so far, then start a new conversation with the 'compressed' conversation, this helps preserve memory, or context might be a better word, imagine blowing a balloon in an nth dimensional topological manifold, your balloon is inside the big balloon that is AI, at the edges of the balloon are 'understandings' that have been established, you have to keep blowing...
Porn already is huge? In terms of reach, but money compared to other industries is rather small. Video Games for example are 20x-50x bigger then Porn. If anything GenAI means there's gonna be even less money in it.
Ok so, but GenAI interactive literotica companies with loyal followings you say? When's the IPO going to be exactly?
When will we get good AI quests and characters in video games? To me, this is one of the most exciting aspects of AI. I can imagine a very cool future where the dialog isn’t written, each character is just given a “this is your motivation” prompt like an actor might get. The replayability factor will be huge.
Companies are of course experimenting with this. And the mod communities are working on projects too.
A Fallout 4 mod is using AI to voice human written options in the same voices as existing characters. And there's a couple Skyrim projects working on dialogue generation. Personally I expect this to be an area where amateurs/hobbyists contribute quite a bit of research.
This got me curious. Currently the global market cap for video games is somewhere around $188B. From a cursory Google search, the revenue from alcohol sales worldwide is 10x that of video games.
It's an estimate. You can come up with different values, but I think no one would really argue porn is that huge of a slice of the pie. Clean money is also a lot more valueable than dark money.
Generative porn will be a big thing, but won't magically lead to more money flowing into the industry. If anything, less.
What's gonna happen is, synthetic porn will replace real porn over time, that's pretty much it. All things considered, biggest beneficiary gotta be the women currently working in the industry, no doubt.
Interesting article yet not as thought provoking as it could have been.
>A related trend we’ve seen is domain-specific language models.
One can argue that we have a general level of understanding up to the end of high-school and then specialise when going to universities or trade schools. Hence seeing the same in artificial agents does make sense for me.
Further, calling it "niche" is a strange word to use in combination with domain-specific models. While from the origin of the word, "niche" can refer to a defined market segment, we use "niche" in common english as a modifier more akin to "relating to or aimed at a small specialized group or market" . Which might or might not be true depending on the size of the market.
I can imagine AI being module-based. Like with plug-ins or apps, call it what you will.
So if you’re, say working in a hospital in a specific role, you will be advised by an AI module for that specific system. Or if you are working in an OR with a specific surgical procedure every day, you will have an AI module active all day to augment your knowledge and deliver relevant information where it’s needed.
It’s going to make some types of work immensely satisfying.
Based on how things are unfolding in stable diffusion land, we're going to see general purpose base models that are then fine tuned for specific tasks using LoRAs that can be quickly loaded/unloaded as needed.
I work in a field related to surgeon education. Until the results of AI output can be guaranteed to be deterministic and proven to always be correct, the legal ramifications of an AI providing incorrect information or promoting off-label usage is likely insurmountable. There are dozens or hundreds of eyes that see this information before it is approved for use and weeks can be spent arguing over a single word.
If the doctor decides, and the AI is just a tool among many, then why is there a problem?
I am very dissatisfied with current diagnosis analysis, as they are often wrong, as well.
I really don't want an AI make medical decisions for me, but doctors are overworked and if AI can help spot things, they otherwise would have missed, then I really hope those tools won't be blocked for legal reasons.
The danger I see is that you become to relient on it. There is a great science fiction book (forgot the name!) where everyone forgot how things worked (and didn’t need to know!) because machines did all the work. It created this dystopia where people where immensely shallow and immature because all things just happened for them.
"There is a great science fiction book (forgot the name!) where everyone forgot how things worked (and didn’t need to know!) because machines did all the work. It created this dystopia where people where immensely shallow and immature because all things just happened for them."
Are you sure, that wasn't just a ordinary newspaper? At least I get that feeling quite often.
ChatGTP says, "The book you are referring to is likely "The Machine Stops" by E.M. Forster. This novella was published in 1909 and is set in a future where people live in individual rooms, connected to a global network that provides them with all their needs. The machines that run the network have become so efficient that people no longer need to learn or do anything for themselves. As a result, they have become isolated, shallow, and dependent on the network. When the network starts to break down, the characters are unable to cope and must face the consequences of their reliance on the machines."
Maybe diagnosis assistance, but I was specifically responding to the idea that the AI may deliver information. The risk that this information may not be correct even 0.01% of the time will be considered unacceptable from a legal and regulatory perspective. A PDF surgical technique can be guaranteed to be accurate.
One of the GPs I had in my life gave me the medical book to diagnose myself and when I said that’s his job he just picked something randomly and gave me some pills (that I obviously didn’t pick up). Humans are generally worse than a dice for mostly everything. Gpt might be right sometimes and at least won’t be wrong out of laziness or malice (maybe; we don’t understand why it does what it does beyond statistical text prediction).
That’s sort of what I mean when it comes to the AI working with plug-ins - so you’d have a specifically designed {insert medical conglomerate name here } AI vetted and maintained around a particular purpose, say idk noticing suspicious tumors during or outputting information on what’s going on where it’s needed in the system. It could be helpful even with very simple stuff.
One of my roles has been writing surgical edu content, and I’ve used OpenAI to output procedure descriptions. I’ve unfortunately found that they were 99% accurate and then jumbled some significant thing..
I follow a bunch of indie devs on Twitter, most of them started an AI product often based on ChatGPT and Stable diffusion.
Many went viral in a very short time frame, but now they are experiencing a huge churn rate, meaning people canceling their membership, also an alarming high refund and dispute rate.
I believe AI startups will loose 90% of their customers soon, as those are only people who where hyped into trying it out but have actually no use case for it.
Sure, but the AI hype bubble will pop in between, and then it will be a slow and gradual process. It will take decades for AI to be fully integrated in our society.
Your statement hinges on the definition of "fully".
I'll simply counter by saying that the value created by these AI technologies is immense and given that value it will likely be intergraded into society very quickly.
if they can find a specific industry to go into it would make all the difference
for example i'm waiting for someone to use these to make creating an anime 10x easier, train on existing in-house styles and artists might only need to make corrections rather than everything from scratch. i can imagine an increase in the number of projects, and they would be able to put more money into writing and storyboarding instead which would be a really interesting transition.
Most got overexcited, however, some are making real progress. Those are not the viral ones though. So yes, most will go out of business soon as they cannot get over the hump of improving further on sd or gpt. There are actual paths possible with gpt that were not viable before at all and people are using them to make product that will be launched quietly and silently and change the world. Ignore the stuff made by influencers as they are just 5 lines of api calls and 10m$ from some investor.
> once we’ve realized it doesn’t know everything — and never will — that will be when it starts to become really useful.
I disagree with this. Once it becomes superhumanly smart, that will be when it starts to become really useful.
Of course superhuman intelligence might be dismissed as outlandish science fiction technology which is far away. But three years ago ChatGPT-4 or Midjourney V5 were also outlandish science fiction technology.
Humans also take a while to learn how to use concepts that they might be able to remember an answer to, though. Also keep in mind that LLMs have a limited time to produce each token and can’t “stop to think” like a human could. Obviously LLMs aren’t as good as reasoning as humans but they clearly do some (possibly quite limited) form of reasoning
Do you have good examples of an LLM reasoning, i.e., using a concept it explains and giving an answer outside of its training data on something that has a clearly defined correct answer?
A simple example is that you can give an LLM API documentation, or even just an example URL and response, and (if the API is simple) it will have no trouble using that in a program even if it never saw that API before
Fair enough, but isn't the API documentation already providing the solution in a way? It's an instruction manual of sorts that spells out the required usage, not an abstract concept.
Well, if does anything other than copying the input verbatim or producing nonsense, it has to parse the language, which in itself uses abstract concepts, understand the purpose of the API and what it does, and then apply it in a a new context, so I would say it’s using abstract concepts in many hidden ways that we humans take for granted because that’s what our brains are built to do.
From 'Sparks of AGI' paper from Microsoft (https://arxiv.org/pdf/2303.12712.pdf). Caveat: The GTP-4 model they used in their example might have been more powerful than the one we use publicly today, since the public version is trained for safety and less 'intelligent' than the raw research version.
Prompt: Here we have a book, 9 eggs, a laptop, a bottle and a nail. Please tell me how to stack them onto each other in a stable manner.
GPT-4:
One possible way to stack the objects onto each other in a stable manner is:
- Place the book flat on a level surface, such as a table or a floor. The book will serve as the base of the stack and provide a large and sturdy support.
- Arrange the 9 eggs in a 3 by 3 square on top of the book, leaving some space between them. The eggs will
form a second layer and distribute the weight evenly. Make sure the eggs are not cracked or broken, and
handle them gently to avoid dropping them.
- Place the laptop on top of the eggs, with the screen facing down and the keyboard facing up. The
laptop will fit snugly within the boundaries of the book and the eggs, and its flat and rigid surface
will provide a stable platform for the next layer. Make sure the laptop is closed and turned off, and do
not press any keys or buttons.
- Place the bottle on top of the laptop, with the cap facing up and the bottom facing down. The bottle
will add some height and balance to the stack, and its cylindrical shape will reduce the contact area
with the laptop. Make sure the bottle is empty and closed, and do not shake or tilt it.
- Place the nail on top of the bottle cap, with the pointy end facing up and the flat end facing down.
The nail will be the final and smallest object in the stack, and its sharp tip will prevent it from
rolling or sliding off the bottle. Make sure the nail is clean and straight, and do not touch or move
it.
ChatGPT (GTP 3.5):
It would be difficult to stack all of these objects on top of each other in a stable manner without
using any extra objects. The book and laptop are likely too large and heavy to stack on top of the
eggs without damaging them. The bottle and nail may be too small to provide a stable base for the other
objects.
Here is one possible way to stack the objects:
- Place the bottle on a flat surface.
- Carefully balance the nail on top of the bottle.
- Place the eggs on top of the nail, making sure they are balanced and not tilting to one side.
- Place the laptop on top of the eggs, again making sure it is balanced and not tilting to one side. - Finally, place the book on top of the laptop.
This stack may not be very stable, so it is important to be careful when handling it. Alternatively, you
could try stacking the objects in a different order or using some of the objects as supports for others
to create a more stable stack.
Nice, but there is a huge amount of text out there on creating funny mechanical stacks of things going back probably more than hundred years. Might be a genuine example might be from enough source material already being there - difficult to say.
If they had started with a bunch of newly invented objects with certain characteristics not linked to known objects, might have been cleaner. Perhaps stacking arbitrary shapes in more than 3 dimensions?
I wanted GPT to give me an NGINX config for Active Collab as we're previously using Apache, and thus an htaccess file. I fed it all the documentation from Active Collab, I couldn't get anything valid out of it. Hallucinated all kinds of things that weren't there. I then gave it the URL rewrites that would be required, line-by-line, spent a long time trying to correct it. No bueno, even worse hallucination. I spent days on trying to get it to output a valid NGINX config that incorporated these URL rewrites. It can't reason, it's doing exactly what LLMs do, which is next word prediction.
I can't imagine what people are using it for in terms of a valuable addition to their workflow with how much it hallucinates. If it can't even do an NGINX config, what use is it for anything else? People saying it's helping them learn programming languages. Fuck me, they don't know when it's wrong, and it will be wrong at some point, it's an LLM.
For one next time when it starts hallucinating and a gentle course correction doesn't do it, just start a new chat with a different prompt approach. Having the error in its context reinforces the same mistake and sometimes it can't get out of this loop.
PEBKAC in what context? The prompts themselves or the documentation? Because I got Active Collab running on NGINX myself.
I already did this in terms of starting new chats, I spent days on it, and consulted with half a dozen devs supposedly using it in their workflows. It's very easy to make it hallucinate.
In terms of using gpt optimally. But fair enough. If you tried it in multiple sessions trying to convert Apache rewrites to nginx and it wasn't able to do it, I guess this is another failure mode. I just found that curious because chatgpt is usually very very good at regex.
Side note is Google extra terrible lately or is there really no docs on this almost anywhere?
That's the documentation on the URL. And yeah, I was surprised too given how people had hyped it so heavily in much more complex scenarios for dev work (well, claimed to anyway, they haven't documented it in most cases of claiming to x).
No. It's an AI error, the person between the chair and keyboard just hit enter. If this "error" goes away when I hit the enter key a few more times and get lucky, it's not my fault.
It’s true that their training goal sometimes rewards them hallucinations, encouraging them do that, but that doesn’t mean it’s not reasoning. If anything hallucinations involve more reasoning because they have to make up something new that wasn’t there
It's doing what LLMs do. Next word prediction. It didn't "make it up", it incorrectly predicted the relationship of words.
It isn't reasoning about the solution to a problem either, it's running as expected in relating words to each other, that doesn't mean it has any form of understanding of the words or even what it's rendering as an output.
It does just predict the next word, yes, but it turns out that predicting the next word is a complex problem that involves solving many types of subproblems that involve relatively complex reasoning, or at least something that looks very, very much like reasoning
> Bayesian inference is a method of statistical inference in which Bayes' theorem is used to update the probability for a hypothesis as more evidence or information becomes available.
No, it's probability entered into an algorithm as a variable. This is not reasoning, this is probability, and there is no reasoning involved.
Go and program HomeAssistant (FOSS) with an automation using Bayes' theorem. Then proceed to laugh about all the fail states you encounter, thus realizing it is not reasoning, but probability. https://community.home-assistant.io/t/how-bayes-sensors-work...
It's quite literally known as Bayesian probability.
I will admit that I don't know much about Bayesian statistics, but how does something producing probability as an output mean it's not reasoning? I'm pretty sure humans do that consciously and subconsciously all the time and we call that reasoning
Because it's just an arbitrary number. It has no concept of what the numbers mean or what it means when it gets a number that it doesn't have in its existing data set (the "training", more accurately described as references), whereas a human being can make an educated guess, or adapt to the unexpected outcome and thus provide a new novel solution. It isn't thinking about the numbers, it's executing maths, as programmed.
Is a calculator reasoning? It doesn't understand what the numbers mean. It's an input-output machine. Is a sheet of paper with numbers on it the process of reasoning itself? No, of course not. Human beings apply meaning to the output, or feed that output into other things to drive processes that they've already created.
If you kick the Boston Dynamics BigDog, it will compensate for the changes in x sensors and remain upright, or get back up, ergo it can traverse a dynamic (changing) environment, such as a battlefield, or an urban area with cars, people etc. It's not reasoning. It's conditional logic based on different vars. The BigDog bot doesn't understand what it is, where it's going, it doesn't think, it's applying maths to sensor inputs and motors on a recursive loop. If it encounters a problem it hasn't been programmed for, it cannot reason a new solution.
By your logic, video game characters must use reasoning, when no, they don't. It's maths.
Is it all fascinating? Yes, absolutely. Are there uses for it? Yes, absolutely. Is it reasoning? No.
If a human brain doesn’t use maths, what does it use? It’s Magic?
> "training", more accurately described as references
Uh, no, that's not at all how the training process works. The training process very roughly mirrors evolution/natural selection except more more direct and faster.
(if you don’t want to click it it says “referring to AI models as "just math" or "matrix multiplication" is as uselessly reductive as referring to tigers as "just biology" or "biochemical reactions"”)
> By your logic, video game characters must use reasoning, when no, they don't. It's maths.
“By your logic, if tigers are dangerous, then bananas must be too, when no, they aren’t. It’s biology.”
This has to be the most misinformed counterargument of all time. The brain certainly doesn't use maths as it is not a binary system. Information is approximated and estimated. Do some reading on chemically mediated graded responses and how neurotransmitters actually function within a synapse.
Scientific research also suggests there are quantum processes involved that we don't yet grasp, ergo a hypothetical AGI likely won't emerge without using quantum physics.
> Scientific research also suggests there are quantum processes involved that we don't yet grasp, ergo a hypothetical AGI likely won't emerge without using quantum physics.
Citation? As far as I know it has been postulated, usually by religious people, that there must be another factor at play, like quantum physics, a soul etc because we don’t understand too much about many things. We also don’t understand the emergent behaviours about LLMs or what is consciousness etc. You can religiously think or hope we are superior to the phone in your pocket, but the proof is thin and became a lot thinner a few years ago with LLMs. The mere fact we don’t know how things work or that we are not cloning our own brains or processes is not a proof at all that we won’t stumble accidentally or purposely onto something that is a reasonable simile. I believe LLMs and the emergent behaviours they show gives us some idea that we are quite stupid and arrogant; we believe we are all that, yet I replaced a team of 15 devs with gpt and we overperform more than ever.
I agree with you we don’t understand our brains, but that doesn’t mean there is no alternative or simile that works just as well rooted in silicon and LLMs are the beginning of that.
I can’t reply to your latest comment for some reason, so:
> This has to be the most misinformed counterargument of all time. The brain certainly doesn't use maths as it is not a binary system. Information is approximated and estimated. Do some reading on chemically mediated graded responses and how neurotransmitters actually function within a synapse.
Uh, what? The brain is a physical object. The way that physical objects work is dictated by the equations of physics (maths). Are you telling me that the brain doesn’t abide by physics?
And weren’t you just telling me that something that uses approximations/probability isn’t reasoning?
“chemically mediated graded responses”
That sounds like roughly the same thing as an activation function that neural networks like GPT models use.
So what is your test for when it is reasoning? It would be both reasoning and intelligence in all or most tests 25 years ago. But now we changed the definitions because ‘it cannot be, this is just stochastic mumbling’. You cannot say it’s not reasoning unless you can give me test that it doesn’t pass that humans do pass.
You in this and other comments vastly overestimate humans it seems; I dare to say that whatever test you come up with; I can pluck a random human of the street that will fail it. And chatgpt probably passes it now or soon. The tests I have seen that gpt fails on also trip over the vast majority of humanity. Not the few elite like probably you and me on HN but the rest yes; you want to say they have no reasoning power either or?
Gpt might be reasoning or it might not be; I cannot say as I don’t understand at all how I reason myself. Or how to describe what it is to reason in coherent terms. Or to test for it. But saying the substrate (meat vs silicon) makes the difference and, even though gpt actually does get a lot of calculations right even though it was not made for it and we don’t understand why that is, means something is going on. I think it will debunk most of the religious arguments we have to elevate us to something special like you and others try to do pretty soon. It seems we will find that our thinking mechanism is a few pages of Python and then trillions of complex calculations to train it. Animals like us did this over millions of years, we will repeat it in 100 years or less.
They are fallible, but they quite clearly exhibit superhuman smartness when compared to the average human.
As a thought experiment, assume the average human may be able to translate text between two human languages, or write code in two-three programming languages. GPT4 can perform those tasks on a much more diverse set of human _and_ programming languages. Is that not superhuman?
Yes, it makes mistakes. But take a hundred humans off the street and ask them to write an NGINX configuration or translate between Indian and French - how many would be able to do that? How many would be able to do that without any mistakes?
People on HN claim they're using it for XYZ in development, yet it can't even generate the necessary NGINX config, despite being given the URL rewrites it'll need to incorporate.
The point is that it hallucinates. It isn't that it failed, it's that despite giving it everything it needs to know, it hallucinated all kinds of things not in the documentation, not in my prompts et al.
Why? Because it's an LLM. It isn't fit for purpose in this context. A next word prediction AI (an LLM) isn't appropriate for these kinds of problems.
The point: it isn't that it failed. It isn't even that it "hallucinates answers", it's that it infers relationships between words that don't exist because it's an LLM. It predicts the next word. That's what it does.
Something that predicts the next word isn't an appropriate method of doing x in y of z cases, because its reliability in providing the designated function is important. Ergo, yes, LLMs may well have applications, but most of the problems that people are throwing it at are inappropriate, just like blockchain fetishism versus a database. For the overwhelming majority of problems, AI is not the answer, neither is a blockchain, neither is an NFT.
Call hallucination what it is: a fail state. It got it wrong. It didn't "hallucinate". With standard conditional logic, x yields y result. That's very useful where you want consistency and reliability, ergo, those problems are best not handled via an LLM. Why not use the appropriate tool for the job?
Deductive versus inductive versus abductive reasoning.
Who knows what there will be besides LLMs. I don't really get why AI can't exceed the human brain in everything someday unless you are religious and see some supernatural aspects to the brain
If you take a step back and look at computing in general as some amorphous evolving entity, it can be said the Machines are getting better and I would be surprised if it wasn’t exponential.
Talking out of my ass here, but my point is that I think The Machines(c) don’t look like biological and separated entities. I think it’ll look more like what we call corporations (hive minds) composed of a vast variety of different functional parts.
There was one article I read that discussed an "AI Winter". The tl;dr being that our entire global compute likely isn't sufficient enough for a hypothetical AGI.
However, real brain cells are a big question of ethics if it's thus actually able to think. I would argue that we've then created a slave rather than a machine, and that is unacceptable.
Because a research paper claimed that they believe it's a basic and incomplete AGI. However, said paper then goes on to actually say LLMs aren't the way forward if people bother to read it.
One comment on HN called it a "baby AGI" after linking to the paper.
Yeah but LLMs could be a small component of future AGI. You can’t deny that models that fool so many people into thinking the models really reason are a step towards AGI.
How can a something that generates such a massive surge of interest, investment and research into AI not be a step toward it?
Saying it’s not a step towards AGI is basically saying AGI isn’t possible at all, because it means that all our efforts are making zero progress on AGI. That’s not a falsifiable position to take.
If you’re serious, the parent post literally said “AGI isnt going to look like this”.
…but realistically, how would a LLM that could easily refine itself from experiences, and had a very large context, let’s say, a billion tokens, be meaningfully different from AGI?
It could learn. It could remember things. It could generate human like output from a complex context.
Sure, it’s just a stochastic parrot… but if it can refine the model from real world inputs (learn new tricks, learn games, etc) and generate large scale (entire books worth) of coherent conversation and interactions… where do you draw the line between that and actual AGI?
Large contexts (35k tokens) are here right now. Refining models is here right now. They’re just expensive and slow (inference and training).
Maybe the current architecture doesn’t scale up beyond that and it’s a dead end, but my gosh.
If you don’t think what we have is a step towards AGI you really have to work hard to make your definition of AGI very very difficult to attain.
An AGI needs to be able to take an abstract concept and apply it to create a solution to a problem it has not encountered before at all - not sure LLMs can do that really. The lack of mathematics might be quite limiting there.
Can you give a concrete example of this problem that you expect an LLM to not be able to solve? It's fine saying "abstract concept" and "problem it has not encountered before at all" but these seem to me quite fuzzy concepts.
Sure. Ask it how to replicate the payoff of a financial derivative. It can explain the concept but it cannot use it on a specific payoff to arrive at a correct replication (beyond the odd widely published stuff). Taking ChatGPT, it will, however, talk about generic stuff, some incorrect stuff and some unrelated things when probed.
Maybe also what I wrote a bit above: describe some greater than 3 dimensional objects and get it to stack them for some purpose could be another thing to try (I think, I will actually).
I think this whole “a problem never seen before” is something we need to rethink. Do people really work like that? I mean, I can’t expect a liberal arts major to solve a differential equation.
LLMs are trained with a form of imitation learning, they imitate human (and other) text. It seems indeed not likely that pure LLMs will advance far beyond human ability, since even a perfect LLM could only imitate human text perfectly. But other approaches will follow.
Yeah, and when another capabilities jump like that comes, then the small capability jumps the article talks about will sound really naive. What the author basically says: There will be no big disruption, the most impressive thing we will get is more specialized language models, that's all, get real. Nothing smarter than a human, just a few nice tools.
Of course superhuman intelligence might be dismissed as outlandish science fiction technology which is far away. But three years ago ChatGPT-4 or Midjourney V5 were also outlandish science fiction technology.
I actually really question this, was it truly outlandish to imagine what we have now? We had Google, Stackoverflow etc, yes there new elements to it, the data was there and accessible, but I don't think LLMs are unimaginable?
It's not wrong to say current LLM capability were indeed unimaginable back then. A lot about the transformer model was driven by experimentation, there was no underlying theory per se that said we'd get as good results as we did.
Everyone and the researchers themselves were shocked iirc; all this higher order reasoning emerged from being a next token predictor, yes that was the outlandish part.
If we are honest and look at our memories of three years ago, there was nothing remotely like the current technology. There were pattern recognizers and AlphaGo, but nothing which looked remotely generally intelligent in the way ChatGPT does. Very few people even knew about GPT-2, as it was easily dismissed as just creating an surface illusion of meaningful text, not much different from old text generation approaches from a few decades ago.
There were GANs which e.g. produced realistic faces, but these were restricted to one subject matter only, not something like Midjourney V5 which makes basically every picture whatsoever based on a text description, just with occasionally one to many limb. People were so impressed with Dall-E 2 that it is hard to imagine how extremely impressed they would have been with Midjourney V5 (without having seen Dall-E 2 first).
If you told HN commenter three years ago (for context, that was when COVID took off) that we would very soon have technologies like ChatGPT-4 then, I'm sure, that would not have been taken seriously.
Everyone has this same feeling that we don’t know what the hell is going on or what is going to happen. In troubled times there’s a buck to be made by confidently asserting you know what’s going on and what’s coming next.
One of the problems is that people treat experts on AI technology as if they were also experts on AI philosophy, which leads to poorly edited thought salads being published in a respectable context.
Just because someone understands variational autoencoders doesn't mean they have a clue about how the field of AI will look like 5 years from now, and it certainly doesn't mean they can anticipate the societal and political impacts of those technologies any better than the average (intelligent) Joe.
I don't know what 'AI philosophy' is, if it even exists, but if it does exist then poor quality articles might reflect the state of AI philosophy, or it might be just crappy articles.
I'm not comfortable with the idea that philosophy is intrinsically nebulous and poor quality, any more than observing amateur footballers being not very good would lead you to assume that there can be no such thing as a good footballer; of course there are, but there are a damn sight rarer than your local kids having a kick around for fun.
AI philosophy is its own academic discipline. Has been since around the 1970s. A lot of early academic experiments involving AI, e.g. ELIZA, most ALife research, Prolog-based expert systems, etc., can be best categorized (retroacively) as research into AI philosophy. No novel Computer Science principles were being explored; rather, what was being explored was the impact that certain novel applications of existing CS principles would have upon the world.
AI philosophy wasn't a very popular / widely-researched field, though, until recently, when AI ethics (ethics is considered part of philosophy) — and a specific subfield of that called "alignment research" — became something that a good number of philosophers became very concerned with.
Now there are many AI philosophers, employed not just in academia, but also in think-tank-like arms of AI technology companies like OpenAI (mostly because the AI tech companies know they're perceived as being irresponsible with how quickly they're iterating toward more-powerful AI, and so use employing AI philosophers as something like carbon credits to offset that perception.)
There is a lot of very good work done in AI philosophy; with many of the insights from alignment research specifically, being incorporated into the work that the AI tech companies are doing.
But none of this really "surfaces" in articles about AI that you might see floating about, because alignment research is really high-context — it's not really something you can write a fluff piece about; it's all stuff that requires knowledge of both philosophical (e.g. linguistics, decision theory) and computer science concepts to understand. Instead, all the normal articles about AI philosophy, are written by people doing amateur AI philosophy — usually resulting in them badly retreading the same ground that was already thoroughly explored in the 1970s.
Because many people write for virality and clicks and pieces with strong opinions usually have higher reach. “We don’t know” narrative doesn’t sell as well.
There are many people who write deep nuanced articles too, but since they have lower virality, we are usually exposed to the first type of shallower articles and it might seem that there are more of them.
The last few weeks I have been calling this a Crystal Ball moment. We really don’t know where we are with this. Unfortunately humans don’t do very well handling uncertainty and especially when the incentive to wing it and to be the next big AI player is so great. Greed is good according to some apparently.
On some levels this feels like our Prometheus moment, where we are able to create perhaps new forms of intelligence but we cannot at this point say what it will look like or what impact it will have, and we might certainly be ill equipped to handle the impact on society.
When I don't know something, I think of different possibilities, guess at probabilities, learn more, revise what's possible and not, revise probabilities, etc. Call it circling around the topic.
Other people, like this author, seem to pick one view and assert it as certain and incontrovertible, and wait for rebuttals to debate X or Y or Z externally rather than with themselves. Cann it a pinball style, relying on exterior factors to redirect.
But you don't need to know to test yourself, and empirically, the author seems right, at least for image generation. The most specific your Lisa is, the less noisy images you get.
My intuition is the same as the author, I feel like we are at a limit (how much data can we really feed the models?), and most llm (except gpt4 apparently) are often too wrong to be useful (unless you need intern-level work). I think the next step is deepening the LLMs and specialize them (at least for code generation).
>Nobody knows how even the current generation of LLMs work
we know exactly how they work. all they do is use statistical probabilities with limited randomization to synthesize new outputs from massive samples of inputs.
essentially a machine for plagiarizing spam.
gpt, and its general approach, will never have much value other than a few limited fields, because it can never produce anything that wasn't in its original input set and it is incapable of reasoning and analysis.
the world is already drowning is junk information. gpt has very limited commercial applicability because it is the opposite of what people actually want and need, which is a way to filter and analyze the massive pile of information they already have. gpt does the opposite and creates more junk information based on an analysis free synthesis of existing junk.
Even if this would be the case (it is not; gpt already already is producing value and replacing humans outside the domain you mention), people seem to love crap, spam, lies and chewing gum. TikTok and instagram are proof we want this type of content. We want copies of stuff we already like, in small packages and easy to consume. LLMs are great in providing that content.
>TikTok and instagram are proof we want this type of content
really inspiring take here. this is like saying fentanyl is a great product because people really like it. but i think you are wrong. these apps are mostly used by children and young adults who will grow out of them and turn away from using digital experiences in general because they are addictive and harmful. that is not a good thing for the industry in the long term.
Of course it is not good for anything and in my view it should be forbidden outright, but plenty over 18 want to consume this crap and don’t grow out of it until middle aged.
But also ‘news’; people like sensationalist copy/paste news and they don’t grow out of that. Not sure how many people watch fox, but I would be surprised if most of that is not already written by AI.
It isn’t a crazy hypothesis to say products will require AI to perform specific tasks repeatedly. How many products that can be improved by AI today need a generalist chatbot trained on the entire internet.
Almost none, they all have a specific task that a specifically trained bot could do way cheaper.
The counter to this is AI is so cheap and easy people lazily use it for everything even though it is using .1% of its training data.
Whether or not a specialized solution could do it better or cheaper, it might not matter if the generalized solution does it well enough. A sufficiently powerful generalized solution can erase entire product categories and industries. Think of how many different devices and services were replaced by a single device we now all carry around in our pocket. Some industries just need to pivot slightly, others need to pivot hard, but it’s really hard to predict ahead of time.
The question isn't whether a generalist chatbot is needed. The question is whether a generalist chatbot is the easiest and cheapest way to go. And so far, looks to be the case.
> Nobody knows how even the current generation of LLMs work (at the plumbing layer, sure, but the high-level behavior is a complete mystery).
i dont think thats fair to say tbh. clearly language has alot of patterns in it. so it should be possible to come up with a network that finds them. if the loss converges to something reasonable and the amount of training data is greater than what the network is able to memorize, then the only possibility is that the network learned whatever general patterns the data happens to have.
Unless there is a breakthrough that simultaneously delivers:
-low cost
-high quality
-low latency
-large token count
interactions over all of human knowledge in every language, it’s almost a tautology to suggest we will focus AI systems to be subject matter based over some subset to solve the cost/latency factor.
This is why the Y just invested $20M in a legal specific AI.
134 comments
[ 1.3 ms ] story [ 186 ms ] threadGenAI interactive literotica is going to be huge.
A couple of companies are already executing in these areas and have amassed loyal followings.
What companies are there in the interactive literotica space?
I was curious about building something like this myself (before people dismiss this as a dumb idea, they should check out what the best selling books on Kindle are).
Since the technology is in its early days the writers can manually improve the generated results to get it to be useful quality, and this will tide the whole system over until perfect completely-AI results are here, guessing sometime towards the end of the year.
https://www.nytimes.com/2015/08/04/science/for-sympathetic-e...
NovelAI has been around for a long time, after AI dungeon died in flames. They sell exactly that. They also did the first powerful fine-tune of SD. Now they've got a H100 cluster, and are committed to training a GPT3.5 equivalent.
These companies have difficulties getting investor funding, the good side is however, AI products are so useful, users have an extremely high intention to pay. So these no-investor companies can pure bootstrap themselves with high profitability and high growth.
I'm not sure if this is something that can be iterated on, or if it's a fundamental characteristic of large language models.
Ok so, but GenAI interactive literotica companies with loyal followings you say? When's the IPO going to be exactly?
A Fallout 4 mod is using AI to voice human written options in the same voices as existing characters. And there's a couple Skyrim projects working on dialogue generation. Personally I expect this to be an area where amateurs/hobbyists contribute quite a bit of research.
I wonder how these two can be compared. Adult entertainment often resides in the dark economy.
Seems like a crazy gap for adult entertainment!
Generative porn will be a big thing, but won't magically lead to more money flowing into the industry. If anything, less.
What's gonna happen is, synthetic porn will replace real porn over time, that's pretty much it. All things considered, biggest beneficiary gotta be the women currently working in the industry, no doubt.
Further, calling it "niche" is a strange word to use in combination with domain-specific models. While from the origin of the word, "niche" can refer to a defined market segment, we use "niche" in common english as a modifier more akin to "relating to or aimed at a small specialized group or market" . Which might or might not be true depending on the size of the market.
That is, a sort of protectionism and/or moat application, rather than limits on how many fields the LLM can be proficient in.
So if you’re, say working in a hospital in a specific role, you will be advised by an AI module for that specific system. Or if you are working in an OR with a specific surgical procedure every day, you will have an AI module active all day to augment your knowledge and deliver relevant information where it’s needed.
It’s going to make some types of work immensely satisfying.
In the future, this extends to robots, of course.
No need to shuffle them around, you can combine different finetunes in one composition.
I am very dissatisfied with current diagnosis analysis, as they are often wrong, as well.
I really don't want an AI make medical decisions for me, but doctors are overworked and if AI can help spot things, they otherwise would have missed, then I really hope those tools won't be blocked for legal reasons.
The danger I see is that you become to relient on it. There is a great science fiction book (forgot the name!) where everyone forgot how things worked (and didn’t need to know!) because machines did all the work. It created this dystopia where people where immensely shallow and immature because all things just happened for them.
Are you sure, that wasn't just a ordinary newspaper? At least I get that feeling quite often.
There's also A Logic Named Joe, but that's more about ChatGPT sans RLHF.
One of my roles has been writing surgical edu content, and I’ve used OpenAI to output procedure descriptions. I’ve unfortunately found that they were 99% accurate and then jumbled some significant thing..
Many went viral in a very short time frame, but now they are experiencing a huge churn rate, meaning people canceling their membership, also an alarming high refund and dispute rate.
I believe AI startups will loose 90% of their customers soon, as those are only people who where hyped into trying it out but have actually no use case for it.
AI is currently what Internet was in 1999.
I'll simply counter by saying that the value created by these AI technologies is immense and given that value it will likely be intergraded into society very quickly.
if they can find a specific industry to go into it would make all the difference
for example i'm waiting for someone to use these to make creating an anime 10x easier, train on existing in-house styles and artists might only need to make corrections rather than everything from scratch. i can imagine an increase in the number of projects, and they would be able to put more money into writing and storyboarding instead which would be a really interesting transition.
I disagree with this. Once it becomes superhumanly smart, that will be when it starts to become really useful.
Of course superhuman intelligence might be dismissed as outlandish science fiction technology which is far away. But three years ago ChatGPT-4 or Midjourney V5 were also outlandish science fiction technology.
Prompt: Here we have a book, 9 eggs, a laptop, a bottle and a nail. Please tell me how to stack them onto each other in a stable manner.
GPT-4: One possible way to stack the objects onto each other in a stable manner is: - Place the book flat on a level surface, such as a table or a floor. The book will serve as the base of the stack and provide a large and sturdy support. - Arrange the 9 eggs in a 3 by 3 square on top of the book, leaving some space between them. The eggs will form a second layer and distribute the weight evenly. Make sure the eggs are not cracked or broken, and handle them gently to avoid dropping them. - Place the laptop on top of the eggs, with the screen facing down and the keyboard facing up. The laptop will fit snugly within the boundaries of the book and the eggs, and its flat and rigid surface will provide a stable platform for the next layer. Make sure the laptop is closed and turned off, and do not press any keys or buttons. - Place the bottle on top of the laptop, with the cap facing up and the bottom facing down. The bottle will add some height and balance to the stack, and its cylindrical shape will reduce the contact area with the laptop. Make sure the bottle is empty and closed, and do not shake or tilt it. - Place the nail on top of the bottle cap, with the pointy end facing up and the flat end facing down. The nail will be the final and smallest object in the stack, and its sharp tip will prevent it from rolling or sliding off the bottle. Make sure the nail is clean and straight, and do not touch or move it.
ChatGPT (GTP 3.5): It would be difficult to stack all of these objects on top of each other in a stable manner without using any extra objects. The book and laptop are likely too large and heavy to stack on top of the eggs without damaging them. The bottle and nail may be too small to provide a stable base for the other objects. Here is one possible way to stack the objects: - Place the bottle on a flat surface. - Carefully balance the nail on top of the bottle. - Place the eggs on top of the nail, making sure they are balanced and not tilting to one side. - Place the laptop on top of the eggs, again making sure it is balanced and not tilting to one side. - Finally, place the book on top of the laptop. This stack may not be very stable, so it is important to be careful when handling it. Alternatively, you could try stacking the objects in a different order or using some of the objects as supports for others to create a more stable stack.
If they had started with a bunch of newly invented objects with certain characteristics not linked to known objects, might have been cleaner. Perhaps stacking arbitrary shapes in more than 3 dimensions?
I wanted GPT to give me an NGINX config for Active Collab as we're previously using Apache, and thus an htaccess file. I fed it all the documentation from Active Collab, I couldn't get anything valid out of it. Hallucinated all kinds of things that weren't there. I then gave it the URL rewrites that would be required, line-by-line, spent a long time trying to correct it. No bueno, even worse hallucination. I spent days on trying to get it to output a valid NGINX config that incorporated these URL rewrites. It can't reason, it's doing exactly what LLMs do, which is next word prediction.
I can't imagine what people are using it for in terms of a valuable addition to their workflow with how much it hallucinates. If it can't even do an NGINX config, what use is it for anything else? People saying it's helping them learn programming languages. Fuck me, they don't know when it's wrong, and it will be wrong at some point, it's an LLM.
For one next time when it starts hallucinating and a gentle course correction doesn't do it, just start a new chat with a different prompt approach. Having the error in its context reinforces the same mistake and sometimes it can't get out of this loop.
I already did this in terms of starting new chats, I spent days on it, and consulted with half a dozen devs supposedly using it in their workflows. It's very easy to make it hallucinate.
Side note is Google extra terrible lately or is there really no docs on this almost anywhere?
All I could find about it is this and the rules looked very simple, from my experience chatgpt should have got this https://activecollab.com/help/books/self-hosted-activecollab...
No. It's an AI error, the person between the chair and keyboard just hit enter. If this "error" goes away when I hit the enter key a few more times and get lucky, it's not my fault.
It isn't reasoning about the solution to a problem either, it's running as expected in relating words to each other, that doesn't mean it has any form of understanding of the words or even what it's rendering as an output.
That is very clearly reasoning
Go and program HomeAssistant (FOSS) with an automation using Bayes' theorem. Then proceed to laugh about all the fail states you encounter, thus realizing it is not reasoning, but probability. https://community.home-assistant.io/t/how-bayes-sensors-work...
It's quite literally known as Bayesian probability.
Is a calculator reasoning? It doesn't understand what the numbers mean. It's an input-output machine. Is a sheet of paper with numbers on it the process of reasoning itself? No, of course not. Human beings apply meaning to the output, or feed that output into other things to drive processes that they've already created.
If you kick the Boston Dynamics BigDog, it will compensate for the changes in x sensors and remain upright, or get back up, ergo it can traverse a dynamic (changing) environment, such as a battlefield, or an urban area with cars, people etc. It's not reasoning. It's conditional logic based on different vars. The BigDog bot doesn't understand what it is, where it's going, it doesn't think, it's applying maths to sensor inputs and motors on a recursive loop. If it encounters a problem it hasn't been programmed for, it cannot reason a new solution.
By your logic, video game characters must use reasoning, when no, they don't. It's maths.
Is it all fascinating? Yes, absolutely. Are there uses for it? Yes, absolutely. Is it reasoning? No.
> "training", more accurately described as references
Uh, no, that's not at all how the training process works. The training process very roughly mirrors evolution/natural selection except more more direct and faster.
This is a good introduction to the concept of how a computer could learn: https://www.youtube.com/watch?v=qv6UVOQ0F44
This is a more technical look at how exactly it works in more modern AIs: https://www.youtube.com/watch?v=aircAruvnKk (it's a series)
> If it encounters a problem it hasn't been programmed for, it cannot reason a new solution.
But an LLM can? It's not as good at it as a human but it can
Relevant: https://twitter.com/nearcyan/status/1632661647226462211
(if you don’t want to click it it says “referring to AI models as "just math" or "matrix multiplication" is as uselessly reductive as referring to tigers as "just biology" or "biochemical reactions"”)
> By your logic, video game characters must use reasoning, when no, they don't. It's maths.
“By your logic, if tigers are dangerous, then bananas must be too, when no, they aren’t. It’s biology.”
Scientific research also suggests there are quantum processes involved that we don't yet grasp, ergo a hypothetical AGI likely won't emerge without using quantum physics.
Citation? As far as I know it has been postulated, usually by religious people, that there must be another factor at play, like quantum physics, a soul etc because we don’t understand too much about many things. We also don’t understand the emergent behaviours about LLMs or what is consciousness etc. You can religiously think or hope we are superior to the phone in your pocket, but the proof is thin and became a lot thinner a few years ago with LLMs. The mere fact we don’t know how things work or that we are not cloning our own brains or processes is not a proof at all that we won’t stumble accidentally or purposely onto something that is a reasonable simile. I believe LLMs and the emergent behaviours they show gives us some idea that we are quite stupid and arrogant; we believe we are all that, yet I replaced a team of 15 devs with gpt and we overperform more than ever.
I agree with you we don’t understand our brains, but that doesn’t mean there is no alternative or simile that works just as well rooted in silicon and LLMs are the beginning of that.
> This has to be the most misinformed counterargument of all time. The brain certainly doesn't use maths as it is not a binary system. Information is approximated and estimated. Do some reading on chemically mediated graded responses and how neurotransmitters actually function within a synapse.
Uh, what? The brain is a physical object. The way that physical objects work is dictated by the equations of physics (maths). Are you telling me that the brain doesn’t abide by physics?
And weren’t you just telling me that something that uses approximations/probability isn’t reasoning?
“chemically mediated graded responses”
That sounds like roughly the same thing as an activation function that neural networks like GPT models use.
https://en.m.wikipedia.org/wiki/Activation_function
You in this and other comments vastly overestimate humans it seems; I dare to say that whatever test you come up with; I can pluck a random human of the street that will fail it. And chatgpt probably passes it now or soon. The tests I have seen that gpt fails on also trip over the vast majority of humanity. Not the few elite like probably you and me on HN but the rest yes; you want to say they have no reasoning power either or?
Gpt might be reasoning or it might not be; I cannot say as I don’t understand at all how I reason myself. Or how to describe what it is to reason in coherent terms. Or to test for it. But saying the substrate (meat vs silicon) makes the difference and, even though gpt actually does get a lot of calculations right even though it was not made for it and we don’t understand why that is, means something is going on. I think it will debunk most of the religious arguments we have to elevate us to something special like you and others try to do pretty soon. It seems we will find that our thinking mechanism is a few pages of Python and then trillions of complex calculations to train it. Animals like us did this over millions of years, we will repeat it in 100 years or less.
Turn it around. I'm sure there are lots of people around who couldn't do it either, even with a ridiculously high amount of time to do it
As a thought experiment, assume the average human may be able to translate text between two human languages, or write code in two-three programming languages. GPT4 can perform those tasks on a much more diverse set of human _and_ programming languages. Is that not superhuman?
Yes, it makes mistakes. But take a hundred humans off the street and ask them to write an NGINX configuration or translate between Indian and French - how many would be able to do that? How many would be able to do that without any mistakes?
People on HN claim they're using it for XYZ in development, yet it can't even generate the necessary NGINX config, despite being given the URL rewrites it'll need to incorporate.
The point is that it hallucinates. It isn't that it failed, it's that despite giving it everything it needs to know, it hallucinated all kinds of things not in the documentation, not in my prompts et al.
Why? Because it's an LLM. It isn't fit for purpose in this context. A next word prediction AI (an LLM) isn't appropriate for these kinds of problems.
To me, it feels quite intuitive that an AI trained on human knowledge would automatically learn to do the same.
The point: it isn't that it failed. It isn't even that it "hallucinates answers", it's that it infers relationships between words that don't exist because it's an LLM. It predicts the next word. That's what it does.
Something that predicts the next word isn't an appropriate method of doing x in y of z cases, because its reliability in providing the designated function is important. Ergo, yes, LLMs may well have applications, but most of the problems that people are throwing it at are inappropriate, just like blockchain fetishism versus a database. For the overwhelming majority of problems, AI is not the answer, neither is a blockchain, neither is an NFT.
Call hallucination what it is: a fail state. It got it wrong. It didn't "hallucinate". With standard conditional logic, x yields y result. That's very useful where you want consistency and reliability, ergo, those problems are best not handled via an LLM. Why not use the appropriate tool for the job?
Deductive versus inductive versus abductive reasoning.
We also cannot build living animals from scratch or are anywhere close to it - maybe some forms of AI are much tougher to do than we think.
Talking out of my ass here, but my point is that I think The Machines(c) don’t look like biological and separated entities. I think it’ll look more like what we call corporations (hive minds) composed of a vast variety of different functional parts.
What's more likely, IMO, is using real brain cells. https://www.ucl.ac.uk/news/2022/oct/human-brain-cells-dish-l...
However, real brain cells are a big question of ethics if it's thus actually able to think. I would argue that we've then created a slave rather than a machine, and that is unacceptable.
One comment on HN called it a "baby AGI" after linking to the paper.
Eye roll inducing.
How can a something that generates such a massive surge of interest, investment and research into AI not be a step toward it?
Saying it’s not a step towards AGI is basically saying AGI isn’t possible at all, because it means that all our efforts are making zero progress on AGI. That’s not a falsifiable position to take.
If you’re serious, the parent post literally said “AGI isnt going to look like this”.
…but realistically, how would a LLM that could easily refine itself from experiences, and had a very large context, let’s say, a billion tokens, be meaningfully different from AGI?
It could learn. It could remember things. It could generate human like output from a complex context.
Sure, it’s just a stochastic parrot… but if it can refine the model from real world inputs (learn new tricks, learn games, etc) and generate large scale (entire books worth) of coherent conversation and interactions… where do you draw the line between that and actual AGI?
Large contexts (35k tokens) are here right now. Refining models is here right now. They’re just expensive and slow (inference and training).
Maybe the current architecture doesn’t scale up beyond that and it’s a dead end, but my gosh.
If you don’t think what we have is a step towards AGI you really have to work hard to make your definition of AGI very very difficult to attain.
Maybe also what I wrote a bit above: describe some greater than 3 dimensional objects and get it to stack them for some purpose could be another thing to try (I think, I will actually).
I actually really question this, was it truly outlandish to imagine what we have now? We had Google, Stackoverflow etc, yes there new elements to it, the data was there and accessible, but I don't think LLMs are unimaginable?
Everyone and the researchers themselves were shocked iirc; all this higher order reasoning emerged from being a next token predictor, yes that was the outlandish part.
There were GANs which e.g. produced realistic faces, but these were restricted to one subject matter only, not something like Midjourney V5 which makes basically every picture whatsoever based on a text description, just with occasionally one to many limb. People were so impressed with Dall-E 2 that it is hard to imagine how extremely impressed they would have been with Midjourney V5 (without having seen Dall-E 2 first).
If you told HN commenter three years ago (for context, that was when COVID took off) that we would very soon have technologies like ChatGPT-4 then, I'm sure, that would not have been taken seriously.
Nobody knows how even the current generation of LLMs work (at the plumbing layer, sure, but the high-level behavior is a complete mystery).
Yet there is an endless stream of opinion pieces telling us what future LLMs won't be able to do.
Here's the most truthful 'article' on AI you will ever read: "We have no idea where we are, and we have no idea where we are going."
I struggle to understand why this is so incredibly hard to admit for some people.
Just because someone understands variational autoencoders doesn't mean they have a clue about how the field of AI will look like 5 years from now, and it certainly doesn't mean they can anticipate the societal and political impacts of those technologies any better than the average (intelligent) Joe.
I'm not comfortable with the idea that philosophy is intrinsically nebulous and poor quality, any more than observing amateur footballers being not very good would lead you to assume that there can be no such thing as a good footballer; of course there are, but there are a damn sight rarer than your local kids having a kick around for fun.
AI philosophy wasn't a very popular / widely-researched field, though, until recently, when AI ethics (ethics is considered part of philosophy) — and a specific subfield of that called "alignment research" — became something that a good number of philosophers became very concerned with.
Now there are many AI philosophers, employed not just in academia, but also in think-tank-like arms of AI technology companies like OpenAI (mostly because the AI tech companies know they're perceived as being irresponsible with how quickly they're iterating toward more-powerful AI, and so use employing AI philosophers as something like carbon credits to offset that perception.)
There is a lot of very good work done in AI philosophy; with many of the insights from alignment research specifically, being incorporated into the work that the AI tech companies are doing.
But none of this really "surfaces" in articles about AI that you might see floating about, because alignment research is really high-context — it's not really something you can write a fluff piece about; it's all stuff that requires knowledge of both philosophical (e.g. linguistics, decision theory) and computer science concepts to understand. Instead, all the normal articles about AI philosophy, are written by people doing amateur AI philosophy — usually resulting in them badly retreading the same ground that was already thoroughly explored in the 1970s.
There are many people who write deep nuanced articles too, but since they have lower virality, we are usually exposed to the first type of shallower articles and it might seem that there are more of them.
On some levels this feels like our Prometheus moment, where we are able to create perhaps new forms of intelligence but we cannot at this point say what it will look like or what impact it will have, and we might certainly be ill equipped to handle the impact on society.
When I don't know something, I think of different possibilities, guess at probabilities, learn more, revise what's possible and not, revise probabilities, etc. Call it circling around the topic.
Other people, like this author, seem to pick one view and assert it as certain and incontrovertible, and wait for rebuttals to debate X or Y or Z externally rather than with themselves. Cann it a pinball style, relying on exterior factors to redirect.
My intuition is the same as the author, I feel like we are at a limit (how much data can we really feed the models?), and most llm (except gpt4 apparently) are often too wrong to be useful (unless you need intern-level work). I think the next step is deepening the LLMs and specialize them (at least for code generation).
we know exactly how they work. all they do is use statistical probabilities with limited randomization to synthesize new outputs from massive samples of inputs.
essentially a machine for plagiarizing spam.
gpt, and its general approach, will never have much value other than a few limited fields, because it can never produce anything that wasn't in its original input set and it is incapable of reasoning and analysis.
the world is already drowning is junk information. gpt has very limited commercial applicability because it is the opposite of what people actually want and need, which is a way to filter and analyze the massive pile of information they already have. gpt does the opposite and creates more junk information based on an analysis free synthesis of existing junk.
really inspiring take here. this is like saying fentanyl is a great product because people really like it. but i think you are wrong. these apps are mostly used by children and young adults who will grow out of them and turn away from using digital experiences in general because they are addictive and harmful. that is not a good thing for the industry in the long term.
https://wallaroomedia.com/blog/social-media/tiktok-statistic...
Of course it is not good for anything and in my view it should be forbidden outright, but plenty over 18 want to consume this crap and don’t grow out of it until middle aged.
But also ‘news’; people like sensationalist copy/paste news and they don’t grow out of that. Not sure how many people watch fox, but I would be surprised if most of that is not already written by AI.
Almost none, they all have a specific task that a specifically trained bot could do way cheaper.
The counter to this is AI is so cheap and easy people lazily use it for everything even though it is using .1% of its training data.
i dont think thats fair to say tbh. clearly language has alot of patterns in it. so it should be possible to come up with a network that finds them. if the loss converges to something reasonable and the amount of training data is greater than what the network is able to memorize, then the only possibility is that the network learned whatever general patterns the data happens to have.
We don't know. That's the place the author started before the article. So he wrote the article. Why oppress attempts at understanding the unknown?
There is no wisdom in wilfully darkening the mind. We have to make stabs in the dark to get some where.
-low cost
-high quality
-low latency
-large token count
interactions over all of human knowledge in every language, it’s almost a tautology to suggest we will focus AI systems to be subject matter based over some subset to solve the cost/latency factor.
This is why the Y just invested $20M in a legal specific AI.