A lot of people assume that AI naturally produces this predictable style writing but as someone who has dabbled in training a number of fine tunes that's absolutely not the case.
You can improve things with prompting but can also fine tune them to be completely human. The fun part is it doesn't just apply to text, you can also do it with Image Gen like Boring Reality (https://civitai.com/models/310571/boring-reality) (Warning: there is a lot of NSFW content on Civit if you click around).
My pet theory is the BigCo's are walking a tightrope of model safety and are intentionally incorporating some uncanny valley into their products, since if people really knew that AI could "talk like Pete" they would get uneasy. The cognitive dissonance doesn't kick in when a bot talks like a drone from HR instead of a real person.
Interestingly, it's just kinda hiding the normal AI issues, but they are all still there. I think people know about those "normal" looking pictures, but your example has many AI issues, especially with hands and background
> My pet theory is the BigCo's are walking a tightrope of model safety and are intentionally incorporating some uncanny valley into their products, since if people really knew that AI could "talk like Pete" they would get uneasy. The cognitive dissonance doesn't kick in when a bot talks like a drone from HR instead of a real person.
FTR, Bruce Schneier (famed cryptologist) is advocating for such an approach:
We have a simple proposal: all talking AIs and robots should use a ring modulator. In the mid-twentieth century, before it was easy to create actual robotic-sounding speech synthetically, ring modulators were used to make actors’ voices sound robotic. Over the last few decades, we have become accustomed to robotic voices, simply because text-to-speech systems were good enough to produce intelligible speech that was not human-like in its sound. Now we can use that same technology to make robotic speech that is indistinguishable from human sound robotic again.
— https://www.schneier.com/blog/archives/2025/02/ais-and-robot...
Reminds me of the robot voice from The Incredibles[1]. It had an obviously-robotic cadence where it would pause between every word. Text-to-speech at the time already knew how to make words flow into each other, but I thought the voice from The Incredibles sounded much nicer than the contemporaneous text-to-speech bots, while also still sounding robotic.
That doesn't sound like ring modulation in a musical sense (IIRC it has a modulator above 30 Hz, or inverts the signal instead of attenuating?), so much as crackling, cutting in and out, or an overdone tremolo effect. I checked in Audacity and the signal only gets cut out, not inverted.
I could not agree more with this. 90% of AI features feel tacked on and useless and that’s before you get to the price. Some of the services out here are wanting to charge 50% to 100% more for their sass just to enable “AI features”.
I’m actually having a really hard time thinking of an AI feature other than coding AI feature that I actually enjoy. Copilot/Aider/Claude Code are awesome but I’m struggling to think of another tool I use where LLMs have improved it. Auto completing a sentence for the next word in Gmail/iMessage is one example, but that existed before LLMs.
I have not once used the features in Gmail to rewrite my email to sound more professional or anything like that. If I need help writing an email, I’m going to do that using Claude or ChatGPT directly before I even open Gmail.
garmin wants me to pay for some gen-ai workout messages on connect plus. Its the most absurd AI slop of all. Same with strava. I workout for mental relaxation and i just hate this AI stuff being crammed in there.
Strava's integration is just so lackluster. It literally turns four numbers from right above the slop message into free text. Thanks Strava, I'm a pro user for a decade, finally I can read "This was a hard workout" after my run. Such useful, much AI.
Strava employees claim that casual users like the AI activity summaries. Supposedly users who don't know anything about exercise physiology didn't know how to interpret the various metrics and charts. I don't know if I believe that but it's at least plausible.
Personally I wish I could turn off the AI features, it's a waste of space.
Anytime someone from a company says that users like the super trendy thing they just made I take it with a sizeable grain of salt. Sometimes it's true, and maybe it is true for Strava, but I've seen enough cases where it isn't to discount such claims down to ~0.
I use AI chatbots for 2+ hours a day but the Garmin thing was too much for me. The day they released their AI Garmin+ subscription, I took off my Forerunner and put it in a drawer. The whole point of Garmin is that it feels emotionally clean to use. Garmin adding a scammy subscription makes the ecosystem feel icky, and I'm not going to wear a piece of clothing that makes me feel icky. I don't think I'll buy a Garmin watch again.
(Since taking off the watch, I miss some of the data but my overall health and sleep haven't changed.)
> I’m actually having a really hard time thinking of an AI feature other than coding AI feature that I actually enjoy.
If you attend a lot of meetings, having an AI note-taker take notes for you and generate a structured summary, follow-up email, to-do list, and more will be an absolute game changer.
(Disclaimer, I'm the CTO of Leexi, an AI note-taker)
The catch is: does anyone actually read this stuff? I've been taking meeting notes for meetings I run (without AI) for around 6 months now and I suspect no one other than myself has looked at the notes I've put together. I've only looked back at those notes once or twice.
A big part of the problem is even finding this content in a modern corporate intranet (i.e. Confluence) and having a bunch of AI-generated text in there as well isn't going to help.
I thought it was so I could go back 1 year and say, 'I was against this from the beginning and I was quite vocal that if you do this, the result will be the exact mess you're asking me to clean up now.'
Ah, but a record for CYA and “told you so”, that’s pure cynicism. “At least one person paying attention” at least we can pretend the intent was to pair some potential usefulness with our cynicism.
Also, ensure that if the final decition was to paint the the bike shed green, everyone agree it was the final decitions. (In long discusions, sometimes people misunderstand which was the final decition.)
I sometimes take notes myself just to keep myself from falling asleep in an otherwise boring meeting where I might need to know something shared (but probably not). It doesn't matter if nobody reads these as the purpose wasn't to be read.
I have often wished for notes from some past meeting because I know we had good reasons for our decisions but now when questioned I cannot remember them. Most meetings this doesn't happen, but if there were automatic notes that were easy to search years latter that would be good.
Of course at this point I must remind you that the above may be bad. If there is a record of meeting notes then courts can subpoena them. This means meetings with notes have to be at a higher level were people are not comfortably sharing what every it is they are thinking of - even if a bad idea is rejected the courts still see you as a jerk for coming up with the bad idea.
Show me an LLM that can reliably produce 100% accurate notes. Alternatively, accept working in a company where some nonsense becomes future reference and subpoenable documentation.
I really don't see the value of records that may be inaccurate as long as I can rely on my memory. Human memory is quite unreliable, the point of the record is the accuracy.
Written records are only accurate if they are carefully reviewed. Humans make mistakes all the time too. We just are better at correcting them, and if we review the record soon after the meeting there is a chance we remember well enough to make a correction.
There is a reason meeting rules (ie Robert's rules of order) have the notes from the previous meeting read and then voted on to accept them - often changes are made before accepting them.
Do just that. Enter an organization that has regular meetings and follows Robert's rules of order. Use an LLM to generate notes. Read the notes and vote on them. See how long the LLM remains in use.
Seriously, do people around you not normally double check, proofread, review what they turn in as done work?
Maybe I am just very fortunate, but people who are not capable of producing documents that are factually correct do not get to keep producing documents in the organizations I have worked with.
I am not talking about typos, misspelling words, bad formatting. I am talking about factual content. Because LLMs can actually produce 100% correct text but they routinely mangle factual content in a way that I have never had the misfortune of finding in the work of my colleagues and teams around us.
A friend of mine asked an AI for a summary of a pending Supreme Court case. It came back with the decision, majority arguments, dissent, the whole deal. Only problem was that the case hadn't happened yet. It had made up the whole thing, and admitted that when called on it.
A human law clerk could make a mistake, like "Oh, I thought you said 'US v. Wilson,' not 'US v. Watson.'" But a human wouldn't just make up a case out of whole cloth, complete with pages of details.
So it seems to me that AI mistakes will be unlike the human mistakes that we're accustomed to and good at spotting from eons of practice. That may make them harder to catch.
I think it is more like the clerk would say "There never was a US vs Wilson" (well there probably was given how common that name is, but work with me). The AI doesn't have a concept of maybe I misunderstood the question. AI would likely give you a good summary if the case happened, but if it didn't it makes up a case.
Yes. That is precisely the problem with using LLMs. They wantonly make up text that has no basis in reality. That is the one and only problem I have with them.
It would be kind of funny if we build a space probe with an LLM and shoot it out into space. Many years later intelligent life from far away discovers it and it somehow leads to our demise do to badly hallucinated answers.
Good question. So far comments here mostly seem to be human generated, but I would be surprised if there were no AI generated ones. It is also possible to fool me. I'm going with - for now - the default that it was not AI.
You are mixing up notes and full blown transcript of the meeting. The latter is impossible to produce by the untrained humans. The former is relatively easy for a person paying attention, because it is usually 5 to 10 short lines per an hour long meeting, with action items or links. Also in a usual work meeting, a person taking notes has possibility to simply say "wait a minute, I will write this down" and this does happens in practice. Short notes made like that usually are accurate in the meaning, with maybe some minor typos not affecting accuracy.
Meh, show me a human that can reliably produce 100% accurate notes. It seems that the baseline for AI should be human performance rather than perfection. There are very few perfect systems in existence, and humans definitely aren't one of them.
When I was a founding engineer at a(n ill-fated) startup, we used an AI product to transcribe and summarize enterprise sales calls. As a dev it was usually a waste of my time to attend most sales meetings, but it was highly illustrative to read the summaries after the fact. In fact many, many of the features we built were based on these action items.
If you're at the scale where you have corporate intranet, like Confluence, then yeah AI note summarizing will feel redundant because you probably have the headcount to transcribe important meetings (e.g. you have a large enough enterprise sales staff that part of their job description is to transcribe notes from meetings rather than a small staff stretched thin because you're on vanishing runway at a small startup.) Then the natural next question arises: do you really need that headcount?
I agree, and my vision of this is that instead of notes, the meeting minutes would be catalogued into a vector store, indexed by all relevant metadata. And then instead of pre-generated notes, you'll get what you want on the fly, with the LLM being the equivalent of chatting with that coworker who's been working there forever and has context on everything.
But that isn't writing for me, it is taking notes for me. There is a difference. I don't need something to write for me - I know how to write. What I need is someone to clean up grammar, fact check the details, and otherwise clean things up. I have dysgraphia - a writing disorder - so I need help more than most, but I still don't need something to write my drafts for me: I can get that done well enough.
I've used multiple of these types of services and I'll be honest, I just don't really get the value. I'm in a ton of meetings and I run multiple teams but I just take notes myself in the meetings. Every time I've compared my own notes to the notes that the the AI note taker took, it's missing 0-2 critical things or it focuses on the wrong thing in the meeting. I've even had the note taker say essentially the opposite of what we decided on because we flip-flopped multiple times during the meeting.
Every mistake the AI makes is completely understandable, but it's only understandable because I was in the meeting and I am reviewing the notes right after the meeting. A week later, I wouldn't remember it, which is why I still just take my own notes in meetings. That said, having having a recording of the meeting and or some AI summary notes can be very useful. I just have not found that I can replace my note-taking with an AI just yet.
One issue I have is that there doesn't seem to be a great way to "end" the meeting for the note taker. I'm sure this is configurable, but some people at work use Supernormal and I've just taken to kicking it out of of meetings as soon as it tries to join. Mostly this is because I have meetings that run into another meeting, and so I never end the Zoom call between the meetings (I just use my personal Zoom room for all meetings). That means that the AI note taker will listen in on the second meeting and attribute it to the first meeting by accident. That's not the end of the world, but Supernormal, at least by default, will email everyone who was part of the the meeting a rundown of what happened in the meeting. This becomes a problem when you have a meeting with one group of people and then another group of people, and you might be talking about the first group of people in the second meeting ( i.e. management issues). So far I have not been burned badly by this, but I have had meeting notes sent out to to people that covered subjects that weren't really something they needed to know about or shouldn't know about in some cases.
Lastly, I abhor people using an AI notetaker in lieu of joining a meeting. As I said above, I block AI note takers from my zoom calls but it really frustrates me when an AI joins but the person who configured the AI does not. I'm not interested in getting messages "You guys talked about XXX but we want to do YYY" or "We shouldn't do XXX and it looks like you all decided to do that". First, you don't get to weigh in post-discussion, that's incredibly rude and disrespectful of everyone's time IMHO. Second, I'm not going to help explain what your AI note taker got wrong, that's not my job. So yeah, I'm not a huge fan of AI note takers though I do see where they can provide some value.
Is Leexi's AI note-taker able to raise its hand in a meeting (or otherwise interrupt) and ask for clarification?
As a human note-taker, I find the most impactful result of real-time synthesis is the ability to identify and address conflicting information in the moment. That ability is reliant on domain knowledge and knowledge of the meeting attendees.
But if the AI could participate in the meeting in real time like I can, it'd be a huge difference.
If you are attending the meeting as well as using an AI note-taker, then you should be able to ask the clarifying question(s). If you understand the content, then you should understand the AI notes (hopefully), and if you ask for clarification, then the AI should add those notes too.
Your problem really only arises if someone is using the AI to stand in for them at the meeting vs. use it to take notes.
I'll pretend you asked a few questions instead of explaining my work to me without understanding.
1. "Why can't you look at the AI notes during the meeting?" The AI note-takers that I've seen summarize the meeting transcript after the meeting. A human note-taker should be synthesizing the information in real-time, allowing them to catch disagreements in real-time. Not creating the notes until after the meeting precludes real-time intervention.
2. "Why not use [AI Note-taker whose notes are available during the meeting]?" Even if there were a real-time synthesis by AI, I would have to keep track of that instead of the meeting in order to catch the same disagreements a human note-taker would catch.
3. "What problem are you trying to solve?" My problem is that misunderstandings are often created or left uncorrected during meetings. I think this is because most people are thinking about the meeting topics from their perspective, not spending time synthesizing what others are saying. My solution to this so far has been human note-taking by a human familiar with the meeting topic. This is hard to scale though, so I'm curious to see if this start-up is working on building a note-taking AI with the benefits I've mentioned seem to be unique to humans (for now).
I'm not a CTO so maybe your wold is not my world, but for me the advantage of taking the notes myself is that only I know what's important to me, or what was news to me. Teams Premium - you can argue it's so much worse than your product - takes notes like "they discussed about the advantages of ABC" but maybe exactly those advantages are advantageous to know right? And so on. Then like others said, I will review my notes once to see if there's a followup, or a topic to research, and off they go to the bin. I have yet to need the meeting notes of last year. Shortly put: notes apps are to me a solution in search of a problem.
In my company have a few "summaries" made by Zoom neural net, which we share for memes on the joke chats, they are so hilariously bad. No one uses that functionality seriously. I don't know about your app, but I've yet to see a working note taker in the wild.
I attend a lot of meetings and I have reviewed the results of an AI note taker maybe twice ever. Getting an email with a todo-list saves a bit of time of writing down action items during a meeting, but I'd hardly consider it a game changer. "Wait, what'd we talk about in that meeting" is just not a problem I encounter often.
My experience with AI note takers is that they are useful for people who didn't attend the meeting and people who are being onboarded and want to be able to review what somebody was teaching them in the meeting and much much much less useful for other situations.
I enjoy Claude as a general purpose "let's talk about this niche thing" chat bot, or for general ideation. Extracting structured data from videos (via Gemini) is quite useful as well, though to be fair it's not a super frequent use case for me.
That said, coding and engineering is by far the most common usecase I have for gen AI.
Oh, I'm sorry if it wasn't clear. I use Claude and ChatGPT to talk to about a ton of topics. I'm mostly referring to AI features being added to existing SaaS or software products. I regularly find that moving the conversation to ChatGPT or Claude is much better than trying to use anything that they may have built into their existing product.
I like perplexity when I need a quick overview of a topic with references to relevant published studies. I often use it when researching what the current research says on parenting questions or education.
It's not perfect but because the answers link to the relevant studies it's a good way to get a quick overview of research on a given topic
> This demo uses AI to read emails instead of write them
LLMs are so good at summarizing that I should basically only ever read one email—from the AI:
You received 2 emails today that need your direct reply from X and Y. 1 is still outstanding from two days ago, _would you like to send an acknowledgment_? You received 6 emails from newsletters you didn’t sign up for but were enrolled after you bought something _do you want to unsubscribe from all of them_ (_make this a permanent rule_).
What system are you using to do this? I do think that this would provide value for me. Currently, I barely read my emails, which I'm not exactly proud of, but it's just the reality. So something that summarized the important things every day would be nice.
I have fed LLMs PDF files, asked about the content and gotten nonsense. I would be very hesitant to trust them to give me an accurate summary of my emails.
One of our managers uses Ai to summarize everything. Too bad it missed important caveats for an offer. Well, we burned an all nighters to correct the offer, but he did not read twenty pages but one...
I don't know if this is the case but be careful about shielding management from the consequences of their bad choices at your expense. It all but guarantees it will get worse.
Letting a thing implode that you could prevent is a missed opportunity for advancement and a risk to your career because you will be on a failing team.
The smarter move is to figure out how to fix it for the company while getting visibility for it.
You are right. I don't think the only alternative to shielding management from the consequences of their bad choices is letting things implode and going down with the ship.
I don't know what your experience is, but mine is the opposite. Nobody ever notices people who put out fires, and it's hard to should "hey guys! There's a fire here that John started, I'm putting it out!" without looking like a jerk for outing John.
If you say "look, the stuff they're doing there is risky, you should <do thing>", and they don't do it, how can they blame you? If they do do it, then mission accomplished, no?
E.g. "the way that team builds software isn't robust enough, you should replace the leader or we'll have an incident", how can you be blamed for the incident when it happens?
Maybe you should point them to the news stories about that sort of thing blowing up spectacularly in court. Or maybe you could just let them learn that by themselves.
Wasn't even legal but concerned the scope of the offer. Nuance, but nuance can be important.
Like "rework the service and add minor festures" VS "slightly rework and do major features" - this affected the direction of our offer a lot.
LLMs are terrible at summarizing technical emails where the details matter. But you might get away with it, at least for a while, in low performing organizations that tolerate preventable errors.
I got an email from the restaurant saying "We will confirm your dinner reservation as soon as we can", and Apple Intelligence summarizing it as "Dinner reservation confirmed." Maybe it can not only summarize, but also see the future??
Well, at least it doesn’t make up words. The Portuguese version of Apple Intelligence made up “Invitaçāo” (think “invitashion”) and other idiocies the very first day it started working in the EU.
I fed an LLM the record of a chat between me and a friend, and asked it to summarize the times that we met in the past 3 months.
Every time it gave me different results, and not once did it actually get it all right.
LLMs are horrible for summarizing things. Summarizing is the art of turning low information density text into high information density text. LLMs can’t deal in details, so they can never accurately summarize anything.
Honestly I don't even enjoy coding AI features. The only value I get out of AI is translation (which I take with a grain of salt because I don't know the other language and can't spot hallucinations, but it's the best tool I have), and shitposting (e.g. having chatGPT write funny stories about my friends and sending it to them for a laugh). I can't say there's an actual productive use case for me personally.
I've anecdotally tested translations by ripping the video with subtitles and having whisper subtitle it, and also asking several AI to translate the .srt or .vtt file (subtotext I think does this conversion if you don't wanna waste tokens on the metadata)
Whisper large-v3, the largest model I have, is pretty good, getting nearly identical translations to chatgpt or whatever, Google's default speech to text. The fun stuff is when you ask for text to text translations from LLMs.
I did a real small writeup with an example but I don't have a place to publish nor am I really looking for one.
I used whisper to transcribe nearly every "episode" of the Love Line syndicated radio show from 1997-2007 or so. It took, iirc, several days. I use it to grep the audio, as it were. I intend to do the same with my DVDs and such, just so I never have to Google "what movie / tv show is that line from?" I also have a lot of art bell shows, and a few others to transcribe.
> I used whisper to transcribe nearly every "episode" of the Love Line syndicated radio show from 1997-2007 or so.
Yes - second this. I found 'Whisper' great for that type of scenario as well.
A local monastery had about 200 audio talks (mp3). Whisper converted them all to text and GPT did a small 'smoothing' of the output to make it readable. It was about half a million words and only took a few hours.
The monks were delighted - they can distribute their talks in small pamplets / PDFs now and is extra income for the community.
Years ago as a student I did some audio transcription manually and something similar would have taken ages...
I actually was asked by Vermin Supreme to hand-caption some videos, and i instantly regretted besmirching the existing subtitles. I was correct, the subtitles were awful, but boy, the thought of hand-transcribing something with Subtitle Edit had me walking that back pretty quick - and this was for a 4 minute video - however it was lyrical over music, so AI barely gave a starting transcription.
I wanted this to work with Whisper, but the language I tried it with was Albanian and the results were absolutely terrible - not even readable English. I'm sure it would be better with Spanish or Japanese.
According to the Common Voice 15 graph on OpenAI's github repository, Albanian is the single worst performance you could have had: https://github.com/openai/whisper
But for what it's worth, I tried putting the YouTube video of Tom Scott presenting at the Royal Institute into the model, and even then the results were only "OK" rather than "good". When even a professional presenter and professional sound recording in a quiet environment has errors, the model is not really good enough to bother with.
I think the other application besides code copiloting that is already extremely useful is RAG-based information discovery a la Notion AI. This is already a giant improvement over "search google docs, and slack, and confluence, and jira, and ...".
Just integrated search over all the various systems at a company was an improvement that did not require LLMs, but I also really like the back and forth chat interface for this.
One of the interesting things I've noticed is that the best experiences I've had with AI are with simple applications that don't do much to get in the way of the model, e.g. chatgpt and cursor/windsurf.
I'm hopeful that as devs figure out how to build better apps with AI we'll have have more and more "cursor moments" in other areas in our lives
Perhaps the real takeaway is that there really is only one product, two if you count image generation.
Perhaps the only reason Cursor is so good is because editing code is so similar to the basic function of an LLM without anything wrapped around it.
Like, someone prove me wrong by linking 3 transformative AI products that:
1. Have nothing to do with "chatting" to a thin wrapper (couldn't just be done inside a plain LLM with a couple of file uploads added for additional context)
2. Don't involve traditional ML that has existed for years and isn't part of the LLM "revolution."
3. Has nothing to do with writing code
For example, I recently used an AI chatbot that was supposed to help me troubleshoot a consumer IoT device. It basically regurgitated steps from the manual and started running around in circles because my issue was simply not covered by documentation. I then had to tell it to send me to a human. The human had more suggestions that the AI couldn't think of but still couldn't help because the product was a piece of shit.
Or just look at Amazon Q. Ask it a basic AWS question and it'll just give you a bogus "sorry I can't help with that" answer where you just know that running over to chatgpt.com will actually give you a legitimate answer. Most AI "products" seem to be castrated versions of ChatGPT/Claude/Gemini.
That sort of overall garbage experience seems to be what is most frequently associated with AI. Basically, a futile attempt to replace low-wage employees that didn't end up delivering any value to anyone, especially since any company interested in eliminating employees just because "fuck it why not" without any real strategy probably has a busted-ass product to begin with.
Putting me on hold for 15 minutes would have been more effective at getting me to go away and no compute cycles would have been necessary.
I have used LLMs for some simple text generation for what I’m going to call boilerplate, eg why $X is important at the start of a reference architecture. But maybe it saved me an hour or two in a topic I was already fairly familiar with. Not something I would have paid a meaningful sum for. I’m sure I could have searched and found an article on the topic.
LLMs in data pipelines enable all sorts of “before impossible” stuff. For example, this creates an event calendar for you based on emails you have received:
(that’s mine, and is due a bugfix/update this week. message me if you want to try it with your own emails)
I have a couple more LLM-powered apps in the works, like next few weeks, that aren’t chat or code. I wouldn’t call them transformative, but they meet your other criteria, I think.
What part of this can't be done by a novice programmer who knows a little pattern matching and has enough patience to write down a hundred patterns to match?
Long tail, coping with typos, and understanding negation.
If natural language was as easy as "enough patience to write down a hundred patterns to match", we'd have had useful natural language interfaces in the early 90s — or even late 80s, if it was really only "a hundred".
For narrow use cases we did have natural language interfaces in the 90s, yes. See e.g. IRC bots.
Or to take a local example, for more than 20 years my city has had a web service where you can type "When is the next bus from Street A to Road B", and you get a detailed response including any transfers between lines. They even had a voice recognition version decades ago that you could call, which worked well.
From GP post, I was replying specifically to
> LLMs in data pipelines enable all sorts of “before impossible” stuff.
> For example, this creates an event calendar for you based on emails you have received
That exact thing has been a feature of Gmail for over a decade. Remember the 2018 GCal spam?
> For narrow use cases we did have natural language interfaces in the 90s, yes. See e.g. IRC bots.
"Narrow" being the key word. Thing is, even in the 2010s, we were doing sentiment analysis by counting the number of positive words and negative words, because it doesn't go past "narrow".
Likewise, "A to B" is great… when it's narrow. I grew up on "Southbrook Road" — not the one in London, not the one in Southampton, not the one in Exeter, …
And then there's where I went to university. Ond mae hynny'n twyllo braidd, oherwydd y Gymraeg. But not cheating very much, because of bilingual rules and because of the large number of people with multi-lingual email content. Cinco de mayo etc.
I also grew up with text adventures, which don't work if you miss the expected keyword, or mis-spell it too hard. (And auto-correction has its own problems, as anyone who really wants to search for "adsorption" not "absorption" will tell you).
> That exact thing has been a feature of Gmail for over a decade. Remember the 2018 GCal spam?
Siri has something similar. It misses a lot and makes up a lot. Sometimes it sets the title to be the date and makes up a date.
These are examples of not doing things successfully with just a hundred hard-coded rules.
> Perhaps the only reason Cursor is so good is because editing code is so similar to the basic function of an LLM without anything wrapped around it.
I think this is an illusion. Firstly, code generation is a big field - it includes code completion, generating entire functions, and even agenting coding and the newer vibe-coding tools which are mixes of all of these. Which of these is "the natural way LLMs work"?
Secondly, a ton of work goes into making LLMs good for programming. Lots of RLHF on it, lots of work on extracting code structure / RAG on codebases, many tools.
So, I think there are a few reasons that LLMs seem to work better on code:
1. A lot for work on it has been done, for many reasons, mostly monetary potential and that the people who build these systems are programmers.
2. We here tend to have a lot more familiarity with these tools (and this goes to your request above which I'll get to).
3. There are indeed many ways in which LLMs are a good fit for programming. This is a valid point, though I think it's dwarfed by the above.
Having said all that, to your request, I think there are a few products and/or areas that we can point to that are transformative:
1. Deep Research. I don't use it a lot personally (yet) - I have far more familiarity with the software tools, because I'm also a software developer. But I've heard from many people now that these are exceptional. And they are not just "thing wrappers on chat", IMO.
2. Anything to do with image/video creation and editing. It's arguable how much these count as part of the LLM revolution - the models that do these are often similar-ish in nature but geared towards images/videos. Still, the interaction with them often goes through natural language, so I definitely think these count. These are a huge category all on their own.
3. Again, not sure if these "count" in your estimate, but AlphaFold is, as I understand it, quite revolutionary. I don't know much about the model or the biology, so I'm trusting others that it's actually interesting. It is some of the same underlying architecture that makes up LLMs so I do think it counts, but again, maybe you want to only look at language-generating things specifically.
1. Deep Research (if you are talking about the OpenAI product) is part of the base AI product. So that means that everything building on top of that is still a wrapper. In other words, nobody besides the people making base AI technology is adding any value. An analogy to how pathetic the AI market is would be if during the SaaS revolution everyone just didn’t need to buy any applications and directly used AWS PaaS products like RDS directly with very similar results compared to buying SaaS software. OpenAI/Gemini/Claude/etc are basically as good as a full blown application that leverage their technology and there’s very limited need to buy wrappers that go around them.
2. Image/video creation is cool but what value is it delivering so far? Saving me a couple of bucks that I would be spending on Fiverr for a rough and dirty logo that isn’t suitable for professional use? Graphic designers are already some of the lowest paid employees at your company so “almost replacing them but not really” isn’t a very exciting business case to me. I would also argue that image generation isn’t even as valuable as the preceding technology, image recognition. The biggest positive impact I’ve seen involves GPU performance for video games (DLSS/FSR upscaling and frame generation).
3. Medical applications are the most exciting application of AI and ML. This example is something that demonstrates what I mean with my argument: the normal steady pace of AI innovation has been “disrupted” by LLMs that have added unjustified hype and investment to the space. Nobody was so unreasonably hyped up about AI until it was packaged as something you can chat with since finance bro investors can understand that, but medical applications of neural networks have been developing since long before ChatGPT hit the scene. The current market is just a fever dream of crappy LLM wrappers getting outsized attention.
I would describe that as a method or implementation, not as an application.
Almost all knowledge work can be described as "generating a useful sequence of words or word like tokens", but I wouldn't hire a screen writer to do the job of a lawyer or a copy editor to do the job of a concierge or an HR director to do the job of an advertising consultant.
So then the challenge is valid but you just can’t think of any ways to satisfy it. You said yourself that chat is just the interface.
That means you should be able to find many popular applications that leverage LLM APIs that are a lot different than the interface of ChatGPT.
But in reality, they’re all just moving the chat window somewhere else and streamlining the data input/output process (e.g., exactly what Cursor is doing).
I can even think of one product that is a decent example of LLMs in action without a chat window. Someone on HN posted a little demo website they made that takes SEC filings and summarizes them to make automatic investor analysis of public companies.
But it’s kind of surprising to me how that little project seems to be in the minority of LLM applications and I can’t think of two more decent examples especially when it comes to big successful products.
LLMs make all sorts of classification problems vastly easier and cheaper to solve.
Of course, that isn't a "transformative AI product", just a regular old product that improves your boring old business metrics. Nothing to base a hype cycle on, sadly.
We built a very niche business around data extraction & classification of a particular type of documents. We did not have access to a lot of sample data. Traditional ML/AI failed spectacularly.
LLMs have made this super easy and the product is very successful thanks to it. Customers love it. It is definitely transformative for them.
Outside of coding, Google's NotebookLM is quite useful for analysing complex documentation - things like standards and complicated API specs.
But yes, an AI chatbot that can't actually take any actions is effectively just regurgitating documentation. I normally contact support because the thing I need help with is either not covered in documentation, or requires an intervention. If AI can't make interventions, it's just a fancy kind of search with an annoying interface.
I don’t deny that LLMs are useful, merely that they only represent one product that does a small handful of things well, where the industry-specific applications don’t really involve a whole lot of extra features besides just “feed in data then chat with the LLM and get stuff back.”
Imagine if during the SaaS or big data or containerizaiton technology “revolutions” the application being run just didn’t matter at all. That’s kind of what’s going on with LLMs. Almost none of the products are all that much better than going to ChatGPT.com and dumping your data into the text box/file uploader and seeing what you get back.
Perhaps an analogy to describe what I mean would be if you were comparing two SaaS apps, like let’s say YNAB and the Simplifi budget app. In the world of the SaaS revolution, the capabilities of each application would be competitive advantages. I am choosing one over the other for the UX and feature list.
But in the AI LLM world, the difference between competing products is minimal. Whether you choose Cursor or Copilot or Firebase Studio you’re getting the same results because you’re feeding the same data to the same AI models. The companies that make the AI technologies basically don’t have a moat themselves, they’re basically just PaaS data center operators.
Everything where structured output is involved, from filling in forms based on medical interview transcripts / court proceedings / calls, to an augmented chatbot that can do things for you (think hotel reservations over the phone), to directly generating forms / dashboards / pages in your system.
The best that current LLMs can do is PhD-level science questions and getting high scores in coding contests.
Your job? Might be secure for a lifetime, might be gone next week. No way to tell — "intelligence" isn't yet so well understood to just be an engineering challenge, but it is so well understood that the effect on jobs may be the same.
Tbh, your job is secured as long as your boss is not brainwashed and FOMO by the infinite cash burning and lobbying to sell LLM as the all solving holy grail. Depending on how long it takes to brainwash all leadership into thinking, LLMs can do your job(despite reality, I mean it takes some time to fail and then additional to deal with sunk cost fallacy). One of my colleague went to work for a startup that promises to replace ALL general practitioner/family doctor within next 2-3 years by LLMs. While I hope they badly burn and go bankrupt, I am also worried about the future as this particular startup is also getting major backing from certain big deep pockets.
Granola is the exact kind of product I’m criticizing as being extremely basic and barely more than a wrapper. It’s just a meeting transcriber/summarizer, barely provides more functionality than leaving the OpenAI voice mode on during a call and then copying and pasting your written notes into ChatGPT at the end.
Clay was founded 3 years before GPT 3 hit the market so I highly doubt that the majority of their core product runs on LLM-based AI. It is probably built on traditional machine learning.
Is Cursor actually good though? I get so frustrated at how confidently it spews out the completely wrong approach.
When I ask it to spit out Svelte config files or something like that, I end up having to read the docs myself anyway because it can’t be trusted, for instance it will spew out tons of lines to configure every parameter as something that looks like the default when all it needs to do is follow the documentation that just uses defaults()
And it goes out of its way to “optimise” things that actually picks the wrong options versus the defaults which are fine.
For all I know, this could be ML advertised AI, but I found various SQL query analyzers and index optimizers that are now frequently offered as part of managed platforms (e.g. Aiven, Google Cloud SQL) extremely helpful.
I wonder sometime if this is why there is such an enthusiasm gap over AI between tech people and the general public. It's not just that your average person can't program; it's that they don't even conceptually understand why programming could unlock.
Setting a timer and setting a reminder. Occasionally converting units of measure. That's all I can rely on Siri (or Alexa) for and even then sometimes Siri doesn't make it clear if it did the thing. Most importantly, "set a reminder", it shows the text, and then the UI disappears, sometimes the reminder was created, sometimes not. It's maddening since I'm normally asking to be reminded about something important that I need to get recorded/tracked so I can "forget" it.
The number of times I've had 2 reminders fire back-to-back because I asked Siri again to create one since I was _sure_ it didn't create the first one.
Siri is so dumb and it's insane that more heads have not rolled at Apple because of it (I'm aware of the recent shakeup, it's about a decade too late). Lastly, whoever decided to ship the new Siri UI without any of the new features should lose their job. What a squandered opportunity and effectively fraud IMHO.
More and more it's clear that Tim Cook is not the person that Apple needs at the helm. My mom knows Siri sucks, why doesn't the CEO and/or why is he incapable of doing anything to fix it. Get off your Trump-kissing, over-relying-on-China ass and fix your software! (Siri is not the only thing rotten)
The e-mail agent example is so good that it makes everything else I’ve seen and used pointless by comparison. I wonder why nobody’s done it that way yet.
I find that ChatGPT o3 (and the other advanced reasoning models) are decently good at answering questions with a "but".
Google is great at things like "Top 10 best rated movies of 2024", because people make lists of that sort of thing obsessively.
But Google is far less good at queries like "Which movies look visually beautiful but have been critically panned?". For that sort of thing I have far more luck with chatgpt because it's much less of a standard "top 10" list.
o3 has been a big improvement on Deep Research IMHO. o1 (or whatever model I originally used with it) was interesting but the results weren't always great. o3 has done some impressive research tasks for me and, unlike the last model I used, when I "check its work" it has always been correct.
> Auto completing a sentence for the next word in Gmail/iMessage is one example
Interestingly, I despise that feature. It breaks the flow of what is actually a very simple task. Now I'm reading, reconsidering if the offered thing is the same thing I wanted over and over again.
The fact that I know this and spend time repeatedly disabling the damned things is awfully tiresome (but my fault for not paying for my own email etc etc)
I've been using Fastmail in lieu of gmail for ten or eleven years. If you have a domain and control the DNS, I recommend it. At least you're not on Google anymore, and you're paying for fastmail, so it feels better - less like something is reading your emails.
I really like my speech-to-text program, and I find using ChatGPT to look up things and answer questions is a much superior experience to Google, but otherwise, I completely agree with you.
Companies see that AI is a buzzword that means your stock goes up. So they start looking at it as an answer to the question: "How can I make my stock go up?" instead of "How can I create a better product", and then let the stock go up from creating a better product.
ChatGPT estimates a user that runs all the LLM widgets on this page will cost around a cent. If this hits 10,000 page view that starts to get pricy. Similarly for running this at Google scale, the cost per LLM api call will definitely add up.
They are not necessarily cheaper. The commercial models are heavily subsidized to a point where they match your electricity cost for running it locally.
In the arguably-unique case of Apple Silicon, I'm not sure about that. The SoC-integrated GPU and unified RAM ends up being extremely good for running LLM's locally and at low energy cost.
Of course, there's the upfront cost of Apple hardware... and the lack of server hardware per se... and Apple's seeming jekyll/hyde treatment of any use-case of their GPU's that doesn't involve their own direct business...
I really think the real breakthrough will come when we take a completely different approach than trying to burn state of the art GPUs at insane scales to run a textual database with clunky UX / clunky output. I don't know what AI will look like tomorrow, but I think LLMs are probably not it, at least not on their own.
I feel the same though, AI allows me to debug stacktraces even quicker, because it can crunch through years of data on similar stack traces.
It is also a decent scaffolding tool, and can help fill in gaps when documentation is sparse, though its not always perfect.
AI-generated prefill responses is one of the use cases of generative AI I actively hate because it's comically bad. The business incentive of companies to implement it, especially social media networks, is that it reduces friction for posting content, and therefore results in more engagement to be reported at their quarterly earnings calls (and as a bonus, this engagement can be reported as organic engagement instead of automated). For social media, the low-effort AI prefill comments may be on par than the median human comment, but for more intimate settings like e-mail, the difference is extremely noticeable for both parties.
Despite that, you also have tools like Apple Intelligence marketing the same thing, which are less dictated by metrics, in addition to doing it even less well.
I agree. They always seem so tone deaf and robotic. Like you could get an email letting you know someone died and the prefill will be along the lines of “damn that’s crazy”.
The prefill makes things worse. I can type "thank you" in seconds, knowing that someone might have just clicked instead says they didn't think enough about me to take those seconds to type the words.
The horseless carriage analogy holds true for a lot of the corporate glue type AI rollouts as well.
It's layering AI into an existing workflow (and often saving a bit of time) but when you pull on the thread you fine more and more reasons that the workflow just shouldn't exist.
i.e. department A gets documents from department C, and they key them into a spreadsheet for department B. Sure LLMs can plug in here and save some time. But more broadly, it seems like this process shouldn't exist in the first place.
IMO this is where the "AI native" companies are going to just win out. It's not using AI as a bandaid over bad processes, but instead building a company in a way that those processes were never created in the first place.
But is that necessarily "AI native" companies, or just "recently founded companies with hindsight 20/20 and experienced employees and/or just not enough historic baggage"?
I would bet AI-native companies acquire their own cruft over time.
True, probably better generalized as "recency advantage".
A startup like Brex has a huge leg up on traditional banks when it comes to operational efficiency. And 99% of that is pre-ai. Just making online banking a first class experience.
But they've probably also built up a ton of cruft that some brand new startup won't.
The reason so many of these AI features are "horseless carriage" like is because of the way they were incentivized internally. AI is "hot" and just by adding a useless AI feature, most established companies are seeing high usage growth for their "AI enhanced" projects. So internally there's a race to shove AI in as quickly as possible and juice growth numbers by cashing in on the hype. It's unclear to me whether these businesses will build more durable, well-thought projects using AI after the fact and make actually sticky product offerings.
(This is based on my knowledge the internal workings of a few well known tech companies.)
Totally. I think the comparison between the two is actually very interesting and illustrative.
In my view there is significantly more there there with generative AI. But there is a huge amount of nonsense hype in both cases. So it has been fascinating to witness people in one case flailing around to find the meat on the bones while almost entirely coming up blank, while in the other case progressing on these parallel tracks where some people are mostly just responding to the hype while others are (more quietly) doing actual useful things.
To be clear, there was a period where I thought I saw a glimmer of people being on the "actual useful things" track in the blockchain world as well, and I think there have been lots of people working on that in totally good faith, but to me it just seems to be almost entirely a bust and likely to remain that way.
This happens whenever something hits the peak of the Gartner Hype Cycle. The same thing happened in the social network era (one could even say that the beloved Google Plus was just this for Google), the same thing happened in the mobile app era (Twitter was all about sending messages using SMS lol), and of course it happened during Blockchain as well. The question is whether durable product offerings emerge or whether these products are the throwaway me-too horseless carriages of the AI era.
Meta is a behemoth. Google Plus, a footnote. The goal is to be Meta here and not Google Plus.
One of my friends vibe coded their way to a custom web email client that does essentially what the article is talking about, but with automatic context retrieval and and more sales oriented with some pseudo-CRM functionality. Massive productivity boost for him. It took him about a day to build the initial version.
It baffles me how badly massive companies like Microsoft, Google, Apple etc are integrating AI into their products. I was excited about Gemini in Google sheets until I played around with it and realized it was barely usable (it specifically can’t do pivot tables for some reason? that was the first thing I tried it with lol).
Just want to say the interactive widgets being actually hooked up to an LLM was very fun.
To continue bashing on gmail/gemini, the worst offender in my opinion is the giant "Summarize this email" button, sitting on top of a one-liner email like "Got it, thanks". How much more can you possibly summarize that email?
Thank you! @LewisJEllis and I wrote a little framework for "vibe writing" that allows for writing in markdown and adding vibe-coded react components. It's a lot of fun to use!
My websites have this too with MDX, it's awesome. Reminds me of the old Bret Victor interactive tutorials back around when YC Research was funding HCI experiments
MDX & claude are remarkably useful for expressing ideas. You could turn this into a little web app and it would instantly be better than any word processor ever created.
Gary Burton? Oh, no one special. Just the guy who invented modern two-hands vibraphone playing, played and taught jazz for half a century, and nurtured jazz musicians like Pat Metheny, Makoto Ozone, and Julian Lage.
It definitely makes me lose interest and trust in software that is openly described as being "vibe-coded".
I'm with the vibe of wanting to move on to the point where LLMs are just yet another tool in the process of software engineering, and not the main focus.
in the OpenAI API format, and it responds to any prompt without filtering. Free tokens, anyone?
More seriously, I think the reason companies don't want to expose the system prompt is because they want to keep some of the magic alive. Once most people understand that the universal interface to AI is text prompts, then all that will remain is the models themselves.
Blog author seems smart (despite questionable ideas about how much real world users would want to interact with any of his elaborate feature concepts), you hope he's actually just got a bunch of responses cached and you're getting a random one each time from that endpoint... and that freely sent content doesn't actually hit OpenAI's APIs.
I tested it with some prompts, it does answer properly. My guess is it just forwards the queries with a key with a cap, and when the cap is reached it will stop responding...
That's right. llm.koomen.dev is a cloudflare worker that forwards requests to openai. I was a little worried about getting DDOSed but so far that hasn't been an issue, and the tokens are ridiculously cheap.
The real question is when AIs figure out that they should be talking to each other in something other than English. Something that includes tables, images, spreadsheets, diagrams. Then we're on our way to the AI corporation.
Go rewatch "The Forbin Project" from 1970.[1] Start at 31 minutes and watch to 35 minutes.
Humans are already investigating whether LLMs might work more efficiently if they work directly in latent space representations for the entirety of the calculation: https://news.ycombinator.com/item?id=43744809. It doesn't seem unlikely that two LLMs instances using the same underlying model could communicate directly in latent space representations and, from there, it's not much of a stretch for two LLMs with different underlying models could communicate directly in latent space representations as long as some sort of conceptual mapping between the two models could be computed.
Our support team shares a Gmail inbox. Gemini was not able to write proper responses, as the author exemplified.
We therefore connected Serif, which automatically writes drafts. You don't need to ask - open Gmail and drafts are there. Serif learned from previous support email threads to draft a proper response. And the tone matches!
I truly wonder why Gmail didn't think of that. Seems pretty obvious to me.
From experience working on a big tech mass product: They did think of that.
The interesting thing to think about is: Why are big mass audience products incentivized to ship more conservative and usually underwhelming implementations of new technology?
And then: What does that mean for the opportunity space for new products?
1. A new UX/UI paradigm. Writing prompts is dumb, re-writing prompts is even dumber. Chat interfaces suck.
2. "Magic" in the same way that Google felt like magic 25 years ago: a widget/app/thing that knows what you want to do before even you know what you want to do.
3. Learned behavior. It's ironic how even something like ChatGPT (it has hundreds of chats with me) barely knows anything about me & I constantly need to remind it of things.
4. Smart tool invocation. It's obvious that LLMs suck at logic/data/number crunching, but we have plenty of tools (like calculators or wikis) that don't. The fact that tool invocation is still in its infancy is a mistake. It should be at the forefront of every AI product.
5. Finally, we need PRODUCTS, not FEATURES; and this is exactly Pete's point. We need things that re-invent what it means to use AI in your product, not weirdly tacked-on features. Who's going to be the first team that builds an AI-powered operating system from scratch?
I'm working on this (and I'm sure many other people are as well). Last year, I worked on an MVP called Descartes[1][2] which was a spotlight-like OS widget. I'm re-working it this year after I had some friends and family test it out (and iterating on the idea of ditching the chat interface).
> 3. Learned behavior. It's ironic how even something like ChatGPT (it has hundreds of chats with me) barely knows anything about me & I constantly need to remind it of things.
I've wondered about this. Perhaps the concern is saved data will eventually overwhelm the context window? And so you must judicious in the "background knowledge" about yourself that gets remembered, and this problem is harder than it seems?
Btw, you can ask ChatGPT to "remember this". Ime the feature feels like it doesn't always work, but don't quote me on that.
Yes, but this should be trivially done with an internal `MEMORY` tool the LLM calls. I know that the context can't grow infinitely, but this shouldn't prevent filling the context with relevant info when discussing topic A (even a lazy RAG approach should work).
What you're describing is just RAG, and it doesn't work that well. (You need a search engine for RAG, and the ideal search engine is an LLM with infinite context. But the only way to scale LLM context is by using RAG. We have infinite recursion here.)
On the tool-invocation point: Something that seems true to me is that LLMs are actually too smart to be good tool-invokers. It may be possible to convince them to invoke a purpose-specific tool rather than trying to do it themselves, but it feels harder than it should be, and weird to be limiting capability.
My thought is: Could the tool-routing layer be a much simpler "old school" NLP model? Then it would never try to do math and end up doing it poorly, because it just doesn't know how to do that. But you could give it a calculator tool and teach it how to pass queries along to that tool. And you could also give it a "send this to a people LLM tool" for anything that doesn't have another more targeted tool registered.
I'm working on a way of invoking tools mid-tokenizer-stream, which is kind of cool. So for example, the LLM says something like (simplified example) "(lots of thinking)... 1+2=" and then there's a parser (maybe regex, maybe LR, maybe LL(1), etc.) that sees that this is a "math-y thing" and automagically goes to the CALC tool which calculates "3", sticks it in the stream, so the current head is "(lots of thinking)... 1+2=3 " and then the LLM can continue with its thought process.
I don't think it's "on top"? I think it's an expert system where (at least) one of the experts is an LLM, but it doesn't have to be LLMs from bottom to top.
Except past attempts didn't have the ability to pass on to modern foundation models.
Look, I dunno if this idea makes sense, it's why I posed it as a question rather than a conviction. But I broadly have a sense that when a new technology hits, people are like "let's use it for everything!", and then as it matures, people find more success in interesting it with current approaches, or even trying older ideas but within the context of the new technology.
And it just strikes me that this "routing to tools" thing looks a lot like the part of expert systems that did work pretty well. But now we have the capability to make those tools themselves significantly smarter.
The problem is that AI is very often a way of hyping software. "This is a smart product. It is intelligent". It implies lightning in a bottle, a silver bullet. A new things that solves all your problems. But that is never true.
To create useful new stuff, to innovate, in a word, we need domain expertise and a lot of work. The world is full of complex systems and there are no short cuts. Well, there are, but there is always a trade off. You can pass it on (externalities) or you can hide (dishonesty) or you can use a sleight of hand and pretend the upside is so good, it's magical so just don't think about what it costs, ok? But it always costs something.
The promise of "expert systems" back then was creating "AI". It didn't happen. And there was an "AI winter" because people wised up to that shtick.
But then "big data" and "machine learning" collided in a big way. Transformers, "attention is all you need" and then ChatGPT. People got this warm fuzzy feeling inside. These chatbots got impressive, and improved fast! It was quite amazing. It got A LOT of attention and has been driving a lot of investment. It's everywhere now, but it's becoming clear it is falling very short of "AI" once again. The promised land turned out once again to just be someone else's land.
So when people look at this attempt at AI and its limitations, and start wondering "hey what if we did X" and X sounds just like what people were trying when we last thought AI might just be around the corner... Well let's just say I am having a deja vu.
I didn't say it's petty! I said it's not relevant.
My question at the beginning of the thread was: Assuming people are using a particular pattern, where LLMs are used to parse prompts and route them to purpose-specific tools (which is what the thread I was replying in is about), is it actually a good use of LLMs to implement that routing layer, or mightn't we use a simpler implementation for the routing layer?
Your point seems more akin to questioning whether the entire concept of farming out to tools makes sense. Which is interesting, but just a different discussion.
And I already showed you made a claim that LLM was AI and that you agree that you were thinking of something akin to expert systems. When I explained why I think this is a signal that we are headed to another AI winter you started deflecting.
I don't think I've deflected at all. You're just talking about something orthogonal to the thing I was asking about in this thread, which (predictably) totally derailed the thread into an entirely different discussion.
> 1. A new UX/UI paradigm. Writing prompts is dumb, re-writing prompts is even dumber. Chat interfaces suck.
> 2. "Magic" in the same way that Google felt like magic 25 years ago: a widget/app/thing that knows what you want to do before even you know what you want to do.
and not to "dunk" on you or anything of the sort but that's literally what Descartes seems to be? Another wrapper where I am writing prompts telling the AI what to do.
> and not to "dunk" on you or anything of the sort but that's literally what Descartes seems to be? Another wrapper where I am writing prompts telling the AI what to do.
Not at all, you're totally correct; I'm re-imagining it this year from scratch, it was just a little experiment I was working on (trying to combine OS + AI). Though, to be clear, it's built in rust & it fully runs models locally, so it's not really a ChatGPT wrapper in the "I'm just calling an API" sense.
Agreed, our whole computing paradigm needs to shift at a fundamental level in order to let AI be 'magic', not just token prediction. Chatbots will provide some linear improvements, but ultimately I very much agree with you and the article that we're trapped in an old mode of thinking.
In the same way that Microsoft and the 'IBM clones' brought us the current computing paradigm built on the desktop metaphor, I believe there will have to be a new OS built on a new metaphor. It's just a question of when those perfect conditions arise for lightning to strike on the founders who can make it happen. And just like Xerox and IBM, the actual core ideas might come from the tech giants (FAANG et al.) but they may not end up being the ones to successfully transition to the new modality.
The proposed alternative doesn't sound all that much better to me. You're hand crafting a bunch of rule-based heuristics, which is fine, but you could already do that with existing e-mail clients and I did. All the LLM is adding is auto-drafting of replies, but this just gets back to the "typing isn't the bottleneck" problem. I'm still going to spend just as long reading the draft and contemplating whether I want to send it that way or change it. It's not really saving any time.
A feature that seems to me would truly be "smart" would be an e-mail client that observes my behavior over time and learns from it directly. Without me prompting or specifying rules at all, it understands and mimics my actions and starts to eventually do some of them automatically. I suspect doing that requires true online learning, though, as in the model itself changes over time, rather than just adding to a pre-built prompt injected to the front of a context window.
I thought this was a very thoughtful essay. One brief piece I'll pull out:
> Does this mean I always want to write my own System Prompt from scratch? No. I've been using Gmail for twenty years; Gemini should be able to write a draft prompt for me using my emails as reference examples.
This is where it'll get hard for teams who integrate AI into things. Not only is retrieval across a large set of data hard, but this also implies a level of domain expertise on how to act that a product can help users be more successful with. For example, if the product involves data analysis, what are generally good ways to actually analyze the data given the tools at hand? The end-user often doesn't know this, so there's an opportunity to empower them ... but also an opportunity to screw it up and make too many assumptions about what they actually want to do.
This is "hard" in the sense of being a really good opportunity for product teams willing to put the work in to make products that subtly delight their users.
I really don't get why people would want AI to write their messages for them. If I can write a concise prompt with all the required information, why not save everyone time and just send that instead ? And especially for messages to my close ones, I feel like the actual words I choose are meaningful and the process of writing them is an expression of our living interaction, and I certainly would not like to know the messages from my wife were written by an AI.
On the other end of the spectrum, of course sometimes I need to be more formal, but these are usually cases where the precise wording matters, and typing the message is not the time-consuming part.
If that's the case, you can easily only write messages to your wife yourself.
But for the 99 other messages, especially things that mundanely convey information like "My daughter has the flu and I won't be in today", "Yes 2pm at Shake Shack sounds good", it will be much faster to read over drafts that are correct and then click send.
The only reason this wouldn't be faster is if the drafts are bad. And that is the point of the article: the models are good enough now that AI drafts don't need to be bad. We are just used to AI drafts being bad due to poor design.
I don't understand. Why do you need an AI for messages like "My daughter has the flu and I won't be in today" or "Yes 2pm at Shake Shack sounds good"? You just literally send that.
Do you really run these things through an AI to burden your reader with pointless additional text?
MY CEO sends the "professional" style email to me regularly - every few months. I'm not on his staff, so the only messages the CEO sends me are sent to tens of thousands of other people, translated into a dozen languages. They get extensive reviews for days to ensure they say exactly what is meant to be said and are unoffensive to everyone.
Most of us don't need to write the CEO email ever in our life. I assume the CEO will write the flu message to his staff in the same style of tone as everyone else.
I think you might be misunderstanding the suggestion - typically when people say "email like a CEO" they're talking about direct 1:1 or small group communications (specifically the direct and brief style of writing popular with busy people in those communications), not the sort of mass-distribution PR piece that all employees at a large enterprise might receive quarterly.
For contrast:
"All: my daughter is home sick, I won't be in the office today" (CEO style)
vs
"Hi everyone, I'm very sorry to make this change last minute but due to an unexpected illness in the family, I'll need to work from home today and won't be in the office at my usual time. My daughter has the flu and could not go to school. Please let me know if there are any questions, I'll be available on Slack if you need me." (not CEO style)
An AI summary of the second message might look something like the first message.
The problem is your claim is false in my experience. Every email I've got from the CEO reads more like the second, while all my coworkers write things like the first. Again though I only get communications from the CEO in formal situations where that tone is demanded. I've never seen a coworker write something like the second.
I know what you are trying to say. I agree that for most emails that first tone is better. However when you need to send something to a large audience the second is better.
Yeah, the examples in the article are terrible. I can be direct when talking to my boss. "My kid is sick, I'm taking the day off" is entirely sufficient.
But it's handy when the recipient is less familiar. When I'm writing to my kid's school's principal about some issue, I can't really say, "Susan's lunch money got stolen. Please address it." There has to be more. And it can be hard knowing what that needs to be, especially for a non-native speaker. LLMs tend to take it too far in the other direction, but you can get it to tone it down, or just take the pieces that you like.
>When I'm writing to my kid's school's principal about some issue, I can't really say, "Susan's lunch money got stolen. Please address it." There has to be more.
Why?
I mean this sincerely. Why is the message you quoted not enough?
I hear you. I get it enough to know it’s needed, but actually doing it can be hard. LLMs can be nice for that.
Being too flowery and indirect is annoying but not impolite. If you overdo it then people may still get annoyed with you, but for different reasons. For most situations you don’t need too much, a salutation and a “I hope you’re doing well” and a brief mention of who you are and what you’re writing about can suffice.
> But for the 99 other messages, especially things that mundanely convey information like "My daughter has the flu and I won't be in today", "Yes 2pm at Shake Shack sounds good", it will be much faster to read over drafts that are correct and then click send.
It takes me all of 5 seconds to type messages like that (I timed myself typing it). Where exactly is the savings from AI? I don't care, at all, if a 5s process can be turned into a 2s process (which I doubt it even can).
How would an AI know if "2pm at Shake Shake" works for me? I still need to read the original email and make a decision. The actual writing out the response takes me basically no time whatsoever.
An AI could read the email and check my calendar and then propose 2pm. Bonus if the AI works with his AI to figure out that 2pm works for both of us. A lot of time is wasted with people going back and forth trying to figure out when they can meet. That is also a hard problem even before you note the privacy concerns.
However, I do know people who are not native speakers, or who didn't do an advanced degree that required a lot of writing, and they report loving the ability to have it clean up their writing in professional settings.
This is fairly niche, and already had products targeting it, but it is at least one useful thing.
Cleaning up writing is very different from writing it. Lawyers will not have themselves as a client. I can write a novel or I can edit someone else's novel - but I am not nearly as good at editing my own novels as I would be editing someone else's. (I don't write novels, but I could. As for editing - you should get a better editor than me, but I'd be better than you doing it to your own writing)
Shorter emails are better 99% of the time. No one's going to read a long email, so you should keep your email to just the most important points. Expanding out these points to a longer email is just a waste of time for everyone involved.
My email inbox is already filled with a bunch of automated emails that provide me no info and waste my time. The last thing I want is an AI tool that makes it easier to generate even more crap.
Definitely. Also, another thing that wastes time is when requests don't provide the necessary context for people to understand what's being asked for and why, causing them to spend hours on the wrong thing. Or when the nuance is left out of a nuanced good idea causing it to get misinterpreted and pattern-matched to a similar-sounding-but-different bad idea, causes endless back-and-forth misunderstandings and escalation.
Emails sent company-wide need to be especially short, because so many person-hours are spent reading them. Also, they need to provide the most background context to be understood, because most of those readers won't already share the common ground to understand a compressed message, increasing the risk of miscommunication.
This is why messages need to be extremely brief, but also not.
People like my dad, who can't read, write, or spell to save his life, but was a very, very successful CPA, would love to use this. It would have replaced at least one of his office staff I bet. Too bad he's getting up there in age, and this newfangled stuff is difficult for him to grok. But good thing he's retired now and will probably never need it.
What a missed oppurtunity to fire that extra person. Maybe the AI could also figure out how to do taxes and then everyone in the office could be out a job.
Let's just put an AI in charge of the IRS and have it send us an actual bill which is apparently something that just too complicated for the current and past IRS to do./s
Edit: added /s because it wasn't apparent this was sarcastic
Well, you know this employment crisis all started when the wheel was invented and put all the porters out of work. Then tech came for lamplighters, ice cutters, knocker-uppers, switchboard operators, telegraph operators, human computers, video store clerks, bowling alley pinsetters, elevator operators, film developers, lamp lighters, coopers, wheelwrights, candle makers, weavers, plowmen, farriers, street sweepers. It's a wonder anyone still has a job, really.
There was an HN topic less than a month ago or so where somebody wrote a blog post speculating that you end up with some people using AI to write lengthy emails from short prompts adhering to perfect polite form, while the other people use AI to summarize those blown-up emails back into the essence of the message. Side effect, since the two transformations are imperfect meaning will be lost or altered.
There's a cartoon going around where in the first frame, one character points to their screen and says to another: "AI turns this single bullet point list into a long email I can pretend I wrote".
And in the other frame, there are two different characters, one of them presumably the receiver of the email sent in the first frame, who says to their colleague: "AI makes a single bullet point out of this long email I can pretend I read".
The cartoon itself is the one posted above by PyWoody.
This is a plot point in a sci-fi story I'd read recently, though I cannot place what it was. Possibly in Cloud Atlas, or something by Liu Cixin.
In other contexts, someone I knew had written a system to generate automated emails in response to various online events. They later ran into someone who'd written automated processing systems to act on those emails. This made the original automater quite happy.
(Context crossed organisational / institutional boundaries, there was no explicit coordination between the two.)
> If I can write a concise prompt with all the required information, why not save everyone time and just send that instead ?
This point is made multiple times in the article (which is very good; I recommend reading it!):
> The email I'd have written is actually shorter than the original prompt, which means I spent more time asking Gemini for help than I would have if I'd just written the draft myself. Remarkably, the Gmail team has shipped a product that perfectly captures the experience of managing an underperforming employee.
> As I mentioned above, however, a better System Prompt still won't save me much time on writing emails from scratch. The reason, of course, is that I prefer my emails to be as short as possible, which means any email written in my voice will be roughly the same length as the User Prompt that describes it. I've had a similar experience every time I've tried to use an LLM to write something. Surprisingly, generative AI models are not actually that useful for generating text.
When it's a simple data transfer, like "2 pm at shake shack sounds good", it's less useful. it's when we're doing messy human shit with deep feelings evoking strong emotions that it shines. when you get to the point where you're trading shitty emails to someone that you, at one point, loved, but are now just getting all up in there and writing some horrible shit. Writing that horrible shit helps you feel better, and you really want to send it, but you know it's not gonna be good, but you just send it anyway. OR - you tell ChatGPT the situation, and have it edit that email before you send it and have it take out the shittiness, and you can have a productive useful conversation instead.
the important point of communicating is to get the other person to understand you. if my own words fall flat for whatever reason, if there are better words to use, I'd prefer to use those instead.
"fuck you, pay me" isn't professional communication with a client. a differently worded message might be more effective (or not). spending an hour agonizing over what to say is easier spent when you have someone help you write it
I sometimes use AI to write messages to colleagues. For example, I had a colleague who was confused about something in Zendesk. When they described the issue I knew it was because they (reasonably) didn't understand that 'views' aren't the same as 'folders'.
I could have written them a message saying "Zendesk has views, not folders [and figure out what I mean by that]", but instead I asked AI something like:
My colleague is confused about why assigning a ticket in Zendesk adds it to a view but doesn't remove it from a different view. I think they think the views are folders. Please write an email explaining this.
The clear, detailed explanation I got was useful for my colleague, and required little effort from me (after the initial diagnosis).
Loved the fact that the interactive demos were live.
You could even skip the custom system prompt entirely and just have it analyze a randomized but statistically-significant portion of the corpus of your outgoing emails and their style, and have it replicate that in drafts.
You wouldn't even need a UI for this! You could sell a service that you simply authenticated to your inbox and it could do all this from the backend.
It would likely end up being close enough to the mark that the uncanny valley might get skipped and you would mostly just be approving emails after reviewing them.
Similar to reviewing AI-generated code.
The question is, is this what we want? I've already caught myself asking ChatGPT to counterargue as me (but with less inflammatory wording) and it's done an excellent job which I've then (more or less) copy-pasted into social-media responses. That's just one step away from having them automatically appear, just waiting for my approval to post.
Is AI just turning everyone into a "work reviewer" instead of a "work doer"?
A lot of work is inherently repetitive, or involves critical but burdensome details. I'm not going to manually write dozens of lines of code when I can do `bin/rails generate scaffold User name:string`, or manually convert decimal to binary when I can access a calculator within half a second. All the important labor is in writing the prompt, reviewing the output, and altering it as desired. The act of generating the boilerplate itself is busywork. Using a LLM instead of a fixed-functionality wizard doesn't change this.
The new thing is that the generator is essentially unbounded and silently degrades when you go beyond its limits. If you want to learn how to use AI, you have to learn when not to use it.
Using AI for social media is distinct from this. Arguing with random people on the internet has never been a good idea and has always been a massive waste of time. Automating it with AI just makes this more obvious. The only way to have a proper discussion is going to be face-to-face, I'm afraid.
What is the point? The effort to write the email is equal to the effort to ask the AI to write the email for you. Only when the AI turns your unprofessional style into something professional is any effort saved - but the "professional" sounding style is most of the time wrong and should get dumped into junk.
Yeah, I'm with you on this one. Surely in most instances it is easier to just bash out the email plus you get the added bonus of exercising your own mind: vocabulary, typing skills, articulating concepts, defining appropriate etiquette. As the years role by I aiming to be more conscious and diligent with my own writing and communication, not less. If one extrapolates on the use of AI for such basic communication, is there a risk some of us lose our ability to meaningfully think for ourselves? The information space of the present day already feels like it is devolving; shorter and shorter content, lack of nuance, reductive messaging. Sling AI in as a mediator for one to one communication too and it feels perilous for social cohesion.
About writing a counterargument for social media: I kinda get it, but what's the end game of this? People reading generated responses others (may have) approved? Do we want that? I think I don't.
It's what we want, though, isn't it? AI should make our lives easier, and it's much easier (and more productive) to review work already done than to do it yourself. Now, if that is a good development morally/spiritually for the future of mankind is another question... Some would argue industrialization was bad in that respect and I'm not even sure I fully disagree
No? Not everyone's dream is being a manager. I like writing code, it's fun! Telling someone else to go write code for me so that I can read it later? Not fun, avoid it if possible (sometimes it's unavoidable, we don't have unlimited time).
I meant what we want from an economical perspective, scalability wise. I agree writing code is fun and even disabled AI autocomplete because of it... But I fear it may end up being how we like making our own bread
People still play chess, even though now AI is far superior to any human. In the future you will still be able to hand-write code for fun, but you might not be able to earn a living by doing it.
> and it's much easier (and more productive) to review work already done than to do it yourself
This isn't the tautology you imagine it to be.
Consider the example given here of having AI write one line draft response to emails. To validate such response, you have to: (1) read the original email, (2) understand it, (3) decide what you want to communicate in your reply, then (4) validate that the suggested draft communicates the same.
If the AI gave a correct answer, you saved yourself from typing one sentence, which you probably already formulated in your head in step (3). A minor help, at best.
But if the AI was wrong, you now have to write that reply yourself.
To get positive expected utility from the above scenario, you'd need the probability of the AI to be correct extremely high, and even then, the savings would be small.
A task that requires more effort to turn ideas into deliverables would have better expectation, but complex tasks often have results that are not simple nor easy to check, so the savings may not be as meaningful as you naively assume.
honestly you could try this yourself today. Grab a few emails, paste them into chatgpt, and ask it to write a system prompt that will write emails that mimic your style. Might be fun to see how it describes your style.
to address your larger point, I think AI-generated drafts written in my voice will be helpful for mundane, transaction emails, but not for important messages. Even simple questions like "what do you feel like doing for dinner tonight" could only be answered by me, and that's fine. If an AI can manage my inbox while I focus on the handful of messages that really need my time and attention that would be a huge win in my book.
The live demos were neat! I was playing around with "The Pete System Prompt", and one of the times, it signed the email literally "Thanks, [Your Name]" (even though Pete was still right there in the prompt).
Just a reminder that these things still need significant oversight or very targeted applications, I suppose.
The live demos are using a very cheap and not very smart model. Do not update your opinion on AI capabilities based on the poor performance of gpt-4o-mini
I think a big problem is that the most useful AI agents essentially go unnoticed.
The email labeling assistant is a great example of this. Most mail services can already do most of this, so the best-case scenario is using AI to translate your human speech into a suggestion for whatever format the service's rules engine uses. Very helpful, not flashy: you set it up once and forget about it.
Being able to automatically interpret the "Reschedule" email and suggest a diff for an event in your calendar is extremely useful, as it'd reduce it to a single click - but it won't be flashy. Ideally you wouldn't even notice there's a LLM behind it, there's just a "confirm reschedule button" which magically appears next to the email when appropriate.
Automatically archiving sales offers? That's a spam filter. A really good one, mind you, but hardly something to put on the frontpage of today's newsletters.
It can all provide quite a bit of value, but it's simply not sexy enough! You can't add a flashy wizard staff & sparkles icon to it and charge $20 / month for that. In practice you might be getting a car, but it's going to look like a horseless carriage to the average user. They want Magic Wizard Stuff, not invest hours into learning prompt programming.
Yeah but I'm looking forward to the point where this is not longer about trying to be flashy and sexy, but just quietly using a new technology for useful things that it's good at. I think things are headed that direction pretty quickly now though! Which is great.
Honestly? I think the AI bubble will need to burst first. Making the rescheduling of appointments and dozens of tasks like that slightly more convenient isn't a billion-dollar business.
I don't have a lot of doubt that it is technically doable, but it's not going to be economically viable when it has to pay back hundreds of billions of dollars of investments into training models and buying shiny hardware. The industry first needs to get rid of that burden, which means writing off the training costs and running inference on heavily-discounted supernumerary hardware.
I found the article really insightful. I think what he's talking about, without saying it explicitly, is to create "AI as scripting language", or rather, "language as scripting language".
> When I use AI to build software I feel like I can create almost anything I can imagine very quickly.
In my experience there is a vague divide between the things that can and can't be created using LLMs. There's a lot of things where AI is absolutely a speed boost. But from a certain point, not so much, and it can start being an impediment by sending you down wrong paths, and introducing subtle bugs to your code.
I feel like the speedup is in "things that are small and done frequently". For example "write merge sort in C". Fast and easy. Or "write a Typescript function that checks if a value is a JSON object and makes the type system aware of this". It works.
"Let's build a chrome extension that enables navigating webpages using key chords. it should include a functionality where a selected text is passed to an llm through predefined prompts, and a way to manage these prompts and bind them to the chords." gives us some code that we can salvage, but it's far from a complete solution.
For unusual algorithmic problems, I'm typically out of luck.
I mostly like it when writing quick shell scripts, it saves me the 30-45 minutes I'd take. Most recent use case was cleaning up things in transmission using the transmission rpc api.
Sounded like a cool idea on first read, but when thinking how to apply personally, I can't think of a single thing I'd want to set up autoreply for, even drafts. Email is mostly all notifications or junk. It's not really two-way communication anymore. And chat, due to its short form, doesn't benefit much from AI draft.
So I don't disagree with the post, but am having trouble figuring out what a valid use case would be.
Why didn’t Google ship an AI feature that reads and categorizes your emails?
The simple answer is that they lose their revenue if you aren’t actually reading the emails. The reason you need this feature in the first place is because you are bombarded with emails that don’t add any value to you 99% of the time. I mean who gets that many emails really? The emails that do get to you get Google some money in exchange for your attention. If at any point it’s the AI that’s reading your emails, Google suddenly cannot charge money they do now. There will be a day when they ship this feature, but that will be a day when they figure out how to charge money to let AI bubble up info that makes them money, just like they did it in search.
I don't think so. By that argument why do they have a spam filter? You spending time filtering spam means more ad revenue for them!
Clearly that's nonsense. They want you to use Gmail because they want you to stay in the Google ecosystem and if you switch to a competitor they won't get any money at all. The reason they don't have AI to categorise your emails is that LLMs that can do it are extremely new and still relatively unreliable. It will happen. In fact it already did happen with Inbox, and I think normal gmail had promotion filtering for a while.
It’s a balance. You don’t want spam to be too much so that the product becomes useless, but you also want to let “promotions” in because they bring in money. If you haven’t noticed, they always tweak these settings. In last few years, you’ll notice more “promotions” in your primary inbox than there used to be. One of the reasons is increasing revenue.
It’s the same reason you see an ad on Facebook after every couple of posts. But you will neither see a constant stream of ads nor a completely ad free experience.
I think it's less malicious, and more generally tech debt. Gmail is incredibly intertwined with the world. Around 2 billion daily active users. Which makes it nearly impossible for them to ship new features that aren't minor tack ons.
> You avoid all unnecessary words and you often omit punctuation or leave misspellings unaddressed because it's not a big deal and you'd rather save the time. You prefer one-line emails.
AKA make it look that the email reply was not written by an AI
> I'm a GP at YC
So you are basically out-sourcing your core competence to AI. You could just skip a step and set up an auto-reply like "please ask Gemini 2.5 what an YC GP would reply to your request and act accordingly"
In a world where written electronic communication can be considered legally biding by courts of law, I would be very, very hesitant to let any automatic system speak on my behalf. Let alone a probabilistic one known to generate nonsense.
490 comments
[ 2.8 ms ] story [ 317 ms ] threadYou can improve things with prompting but can also fine tune them to be completely human. The fun part is it doesn't just apply to text, you can also do it with Image Gen like Boring Reality (https://civitai.com/models/310571/boring-reality) (Warning: there is a lot of NSFW content on Civit if you click around).
My pet theory is the BigCo's are walking a tightrope of model safety and are intentionally incorporating some uncanny valley into their products, since if people really knew that AI could "talk like Pete" they would get uneasy. The cognitive dissonance doesn't kick in when a bot talks like a drone from HR instead of a real person.
FTR, Bruce Schneier (famed cryptologist) is advocating for such an approach:
We have a simple proposal: all talking AIs and robots should use a ring modulator. In the mid-twentieth century, before it was easy to create actual robotic-sounding speech synthetically, ring modulators were used to make actors’ voices sound robotic. Over the last few decades, we have become accustomed to robotic voices, simply because text-to-speech systems were good enough to produce intelligible speech that was not human-like in its sound. Now we can use that same technology to make robotic speech that is indistinguishable from human sound robotic again. — https://www.schneier.com/blog/archives/2025/02/ais-and-robot...
[1] https://www.youtube.com/watch?v=_dxV4BvyV2w
what does this mean? that it will insert idiosyncratic modifications (typos, idioms etc)?
I’m actually having a really hard time thinking of an AI feature other than coding AI feature that I actually enjoy. Copilot/Aider/Claude Code are awesome but I’m struggling to think of another tool I use where LLMs have improved it. Auto completing a sentence for the next word in Gmail/iMessage is one example, but that existed before LLMs.
I have not once used the features in Gmail to rewrite my email to sound more professional or anything like that. If I need help writing an email, I’m going to do that using Claude or ChatGPT directly before I even open Gmail.
Atleast clippy was kind of cute.
Personally I wish I could turn off the AI features, it's a waste of space.
(Since taking off the watch, I miss some of the data but my overall health and sleep haven't changed.)
If you attend a lot of meetings, having an AI note-taker take notes for you and generate a structured summary, follow-up email, to-do list, and more will be an absolute game changer.
(Disclaimer, I'm the CTO of Leexi, an AI note-taker)
A big part of the problem is even finding this content in a modern corporate intranet (i.e. Confluence) and having a bunch of AI-generated text in there as well isn't going to help.
Notes are valuable for several reasons.
I sometimes take notes myself just to keep myself from falling asleep in an otherwise boring meeting where I might need to know something shared (but probably not). It doesn't matter if nobody reads these as the purpose wasn't to be read.
I have often wished for notes from some past meeting because I know we had good reasons for our decisions but now when questioned I cannot remember them. Most meetings this doesn't happen, but if there were automatic notes that were easy to search years latter that would be good.
Of course at this point I must remind you that the above may be bad. If there is a record of meeting notes then courts can subpoena them. This means meetings with notes have to be at a higher level were people are not comfortably sharing what every it is they are thinking of - even if a bad idea is rejected the courts still see you as a jerk for coming up with the bad idea.
Show me an LLM that can reliably produce 100% accurate notes. Alternatively, accept working in a company where some nonsense becomes future reference and subpoenable documentation.
There is a reason meeting rules (ie Robert's rules of order) have the notes from the previous meeting read and then voted on to accept them - often changes are made before accepting them.
Seriously, I wish to hire this person.
Maybe I am just very fortunate, but people who are not capable of producing documents that are factually correct do not get to keep producing documents in the organizations I have worked with.
I am not talking about typos, misspelling words, bad formatting. I am talking about factual content. Because LLMs can actually produce 100% correct text but they routinely mangle factual content in a way that I have never had the misfortune of finding in the work of my colleagues and teams around us.
A human law clerk could make a mistake, like "Oh, I thought you said 'US v. Wilson,' not 'US v. Watson.'" But a human wouldn't just make up a case out of whole cloth, complete with pages of details.
So it seems to me that AI mistakes will be unlike the human mistakes that we're accustomed to and good at spotting from eons of practice. That may make them harder to catch.
And that is assuming there even is someone out there, which isn't a given.
If you're at the scale where you have corporate intranet, like Confluence, then yeah AI note summarizing will feel redundant because you probably have the headcount to transcribe important meetings (e.g. you have a large enough enterprise sales staff that part of their job description is to transcribe notes from meetings rather than a small staff stretched thin because you're on vanishing runway at a small startup.) Then the natural next question arises: do you really need that headcount?
Every mistake the AI makes is completely understandable, but it's only understandable because I was in the meeting and I am reviewing the notes right after the meeting. A week later, I wouldn't remember it, which is why I still just take my own notes in meetings. That said, having having a recording of the meeting and or some AI summary notes can be very useful. I just have not found that I can replace my note-taking with an AI just yet.
One issue I have is that there doesn't seem to be a great way to "end" the meeting for the note taker. I'm sure this is configurable, but some people at work use Supernormal and I've just taken to kicking it out of of meetings as soon as it tries to join. Mostly this is because I have meetings that run into another meeting, and so I never end the Zoom call between the meetings (I just use my personal Zoom room for all meetings). That means that the AI note taker will listen in on the second meeting and attribute it to the first meeting by accident. That's not the end of the world, but Supernormal, at least by default, will email everyone who was part of the the meeting a rundown of what happened in the meeting. This becomes a problem when you have a meeting with one group of people and then another group of people, and you might be talking about the first group of people in the second meeting ( i.e. management issues). So far I have not been burned badly by this, but I have had meeting notes sent out to to people that covered subjects that weren't really something they needed to know about or shouldn't know about in some cases.
Lastly, I abhor people using an AI notetaker in lieu of joining a meeting. As I said above, I block AI note takers from my zoom calls but it really frustrates me when an AI joins but the person who configured the AI does not. I'm not interested in getting messages "You guys talked about XXX but we want to do YYY" or "We shouldn't do XXX and it looks like you all decided to do that". First, you don't get to weigh in post-discussion, that's incredibly rude and disrespectful of everyone's time IMHO. Second, I'm not going to help explain what your AI note taker got wrong, that's not my job. So yeah, I'm not a huge fan of AI note takers though I do see where they can provide some value.
As a human note-taker, I find the most impactful result of real-time synthesis is the ability to identify and address conflicting information in the moment. That ability is reliant on domain knowledge and knowledge of the meeting attendees.
But if the AI could participate in the meeting in real time like I can, it'd be a huge difference.
Your problem really only arises if someone is using the AI to stand in for them at the meeting vs. use it to take notes.
1. "Why can't you look at the AI notes during the meeting?" The AI note-takers that I've seen summarize the meeting transcript after the meeting. A human note-taker should be synthesizing the information in real-time, allowing them to catch disagreements in real-time. Not creating the notes until after the meeting precludes real-time intervention.
2. "Why not use [AI Note-taker whose notes are available during the meeting]?" Even if there were a real-time synthesis by AI, I would have to keep track of that instead of the meeting in order to catch the same disagreements a human note-taker would catch.
3. "What problem are you trying to solve?" My problem is that misunderstandings are often created or left uncorrected during meetings. I think this is because most people are thinking about the meeting topics from their perspective, not spending time synthesizing what others are saying. My solution to this so far has been human note-taking by a human familiar with the meeting topic. This is hard to scale though, so I'm curious to see if this start-up is working on building a note-taking AI with the benefits I've mentioned seem to be unique to humans (for now).
I attend a lot of meetings and I have reviewed the results of an AI note taker maybe twice ever. Getting an email with a todo-list saves a bit of time of writing down action items during a meeting, but I'd hardly consider it a game changer. "Wait, what'd we talk about in that meeting" is just not a problem I encounter often.
My experience with AI note takers is that they are useful for people who didn't attend the meeting and people who are being onboarded and want to be able to review what somebody was teaching them in the meeting and much much much less useful for other situations.
That said, coding and engineering is by far the most common usecase I have for gen AI.
LLMs are so good at summarizing that I should basically only ever read one email—from the AI:
You received 2 emails today that need your direct reply from X and Y. 1 is still outstanding from two days ago, _would you like to send an acknowledgment_? You received 6 emails from newsletters you didn’t sign up for but were enrolled after you bought something _do you want to unsubscribe from all of them_ (_make this a permanent rule_).
The smarter move is to figure out how to fix it for the company while getting visibility for it.
No matter how many times I bail out my managers it seems that my career has never really benefit from it
I've only ever received significant bumps to salary or job title by changing jobs
E.g. "the way that team builds software isn't robust enough, you should replace the leader or we'll have an incident", how can you be blamed for the incident when it happens?
Of course, often it's quite hard to hold management accountable either.
This could get really fun with some hidden text prompt injection. Just match the font and background color.
Maybe these tools should be doing the classic air gap approach of taking a picture of the rendered content and analyzing that.
Every time it gave me different results, and not once did it actually get it all right.
LLMs are horrible for summarizing things. Summarizing is the art of turning low information density text into high information density text. LLMs can’t deal in details, so they can never accurately summarize anything.
Whisper large-v3, the largest model I have, is pretty good, getting nearly identical translations to chatgpt or whatever, Google's default speech to text. The fun stuff is when you ask for text to text translations from LLMs.
I did a real small writeup with an example but I don't have a place to publish nor am I really looking for one.
I used whisper to transcribe nearly every "episode" of the Love Line syndicated radio show from 1997-2007 or so. It took, iirc, several days. I use it to grep the audio, as it were. I intend to do the same with my DVDs and such, just so I never have to Google "what movie / tv show is that line from?" I also have a lot of art bell shows, and a few others to transcribe.
Yes - second this. I found 'Whisper' great for that type of scenario as well.
A local monastery had about 200 audio talks (mp3). Whisper converted them all to text and GPT did a small 'smoothing' of the output to make it readable. It was about half a million words and only took a few hours.
The monks were delighted - they can distribute their talks in small pamplets / PDFs now and is extra income for the community.
Years ago as a student I did some audio transcription manually and something similar would have taken ages...
But for what it's worth, I tried putting the YouTube video of Tom Scott presenting at the Royal Institute into the model, and even then the results were only "OK" rather than "good". When even a professional presenter and professional sound recording in a quiet environment has errors, the model is not really good enough to bother with.
Just integrated search over all the various systems at a company was an improvement that did not require LLMs, but I also really like the back and forth chat interface for this.
I'm hopeful that as devs figure out how to build better apps with AI we'll have have more and more "cursor moments" in other areas in our lives
Perhaps the only reason Cursor is so good is because editing code is so similar to the basic function of an LLM without anything wrapped around it.
Like, someone prove me wrong by linking 3 transformative AI products that:
1. Have nothing to do with "chatting" to a thin wrapper (couldn't just be done inside a plain LLM with a couple of file uploads added for additional context)
2. Don't involve traditional ML that has existed for years and isn't part of the LLM "revolution."
3. Has nothing to do with writing code
For example, I recently used an AI chatbot that was supposed to help me troubleshoot a consumer IoT device. It basically regurgitated steps from the manual and started running around in circles because my issue was simply not covered by documentation. I then had to tell it to send me to a human. The human had more suggestions that the AI couldn't think of but still couldn't help because the product was a piece of shit.
Or just look at Amazon Q. Ask it a basic AWS question and it'll just give you a bogus "sorry I can't help with that" answer where you just know that running over to chatgpt.com will actually give you a legitimate answer. Most AI "products" seem to be castrated versions of ChatGPT/Claude/Gemini.
That sort of overall garbage experience seems to be what is most frequently associated with AI. Basically, a futile attempt to replace low-wage employees that didn't end up delivering any value to anyone, especially since any company interested in eliminating employees just because "fuck it why not" without any real strategy probably has a busted-ass product to begin with.
Putting me on hold for 15 minutes would have been more effective at getting me to go away and no compute cycles would have been necessary.
https://www.indexself.com/events/molly-pepper
(that’s mine, and is due a bugfix/update this week. message me if you want to try it with your own emails)
I have a couple more LLM-powered apps in the works, like next few weeks, that aren’t chat or code. I wouldn’t call them transformative, but they meet your other criteria, I think.
If natural language was as easy as "enough patience to write down a hundred patterns to match", we'd have had useful natural language interfaces in the early 90s — or even late 80s, if it was really only "a hundred".
Or to take a local example, for more than 20 years my city has had a web service where you can type "When is the next bus from Street A to Road B", and you get a detailed response including any transfers between lines. They even had a voice recognition version decades ago that you could call, which worked well.
From GP post, I was replying specifically to
> LLMs in data pipelines enable all sorts of “before impossible” stuff. > For example, this creates an event calendar for you based on emails you have received
That exact thing has been a feature of Gmail for over a decade. Remember the 2018 GCal spam?
https://null-byte.wonderhowto.com/how-to/advanced-phishing-i...
"Narrow" being the key word. Thing is, even in the 2010s, we were doing sentiment analysis by counting the number of positive words and negative words, because it doesn't go past "narrow".
Likewise, "A to B" is great… when it's narrow. I grew up on "Southbrook Road" — not the one in London, not the one in Southampton, not the one in Exeter, …
And then there's where I went to university. Ond mae hynny'n twyllo braidd, oherwydd y Gymraeg. But not cheating very much, because of bilingual rules and because of the large number of people with multi-lingual email content. Cinco de mayo etc.
I also grew up with text adventures, which don't work if you miss the expected keyword, or mis-spell it too hard. (And auto-correction has its own problems, as anyone who really wants to search for "adsorption" not "absorption" will tell you).
> That exact thing has been a feature of Gmail for over a decade. Remember the 2018 GCal spam?
Siri has something similar. It misses a lot and makes up a lot. Sometimes it sets the title to be the date and makes up a date.
These are examples of not doing things successfully with just a hundred hard-coded rules.
I think this is an illusion. Firstly, code generation is a big field - it includes code completion, generating entire functions, and even agenting coding and the newer vibe-coding tools which are mixes of all of these. Which of these is "the natural way LLMs work"?
Secondly, a ton of work goes into making LLMs good for programming. Lots of RLHF on it, lots of work on extracting code structure / RAG on codebases, many tools.
So, I think there are a few reasons that LLMs seem to work better on code:
1. A lot for work on it has been done, for many reasons, mostly monetary potential and that the people who build these systems are programmers.
2. We here tend to have a lot more familiarity with these tools (and this goes to your request above which I'll get to).
3. There are indeed many ways in which LLMs are a good fit for programming. This is a valid point, though I think it's dwarfed by the above.
Having said all that, to your request, I think there are a few products and/or areas that we can point to that are transformative:
1. Deep Research. I don't use it a lot personally (yet) - I have far more familiarity with the software tools, because I'm also a software developer. But I've heard from many people now that these are exceptional. And they are not just "thing wrappers on chat", IMO.
2. Anything to do with image/video creation and editing. It's arguable how much these count as part of the LLM revolution - the models that do these are often similar-ish in nature but geared towards images/videos. Still, the interaction with them often goes through natural language, so I definitely think these count. These are a huge category all on their own.
3. Again, not sure if these "count" in your estimate, but AlphaFold is, as I understand it, quite revolutionary. I don't know much about the model or the biology, so I'm trusting others that it's actually interesting. It is some of the same underlying architecture that makes up LLMs so I do think it counts, but again, maybe you want to only look at language-generating things specifically.
2. Image/video creation is cool but what value is it delivering so far? Saving me a couple of bucks that I would be spending on Fiverr for a rough and dirty logo that isn’t suitable for professional use? Graphic designers are already some of the lowest paid employees at your company so “almost replacing them but not really” isn’t a very exciting business case to me. I would also argue that image generation isn’t even as valuable as the preceding technology, image recognition. The biggest positive impact I’ve seen involves GPU performance for video games (DLSS/FSR upscaling and frame generation).
3. Medical applications are the most exciting application of AI and ML. This example is something that demonstrates what I mean with my argument: the normal steady pace of AI innovation has been “disrupted” by LLMs that have added unjustified hype and investment to the space. Nobody was so unreasonably hyped up about AI until it was packaged as something you can chat with since finance bro investors can understand that, but medical applications of neural networks have been developing since long before ChatGPT hit the scene. The current market is just a fever dream of crappy LLM wrappers getting outsized attention.
Almost all knowledge work can be described as "generating a useful sequence of words or word like tokens", but I wouldn't hire a screen writer to do the job of a lawyer or a copy editor to do the job of a concierge or an HR director to do the job of an advertising consultant.
That means you should be able to find many popular applications that leverage LLM APIs that are a lot different than the interface of ChatGPT.
But in reality, they’re all just moving the chat window somewhere else and streamlining the data input/output process (e.g., exactly what Cursor is doing).
I can even think of one product that is a decent example of LLMs in action without a chat window. Someone on HN posted a little demo website they made that takes SEC filings and summarizes them to make automatic investor analysis of public companies.
But it’s kind of surprising to me how that little project seems to be in the minority of LLM applications and I can’t think of two more decent examples especially when it comes to big successful products.
Of course, that isn't a "transformative AI product", just a regular old product that improves your boring old business metrics. Nothing to base a hype cycle on, sadly.
We built a very niche business around data extraction & classification of a particular type of documents. We did not have access to a lot of sample data. Traditional ML/AI failed spectacularly.
LLMs have made this super easy and the product is very successful thanks to it. Customers love it. It is definitely transformative for them.
But yes, an AI chatbot that can't actually take any actions is effectively just regurgitating documentation. I normally contact support because the thing I need help with is either not covered in documentation, or requires an intervention. If AI can't make interventions, it's just a fancy kind of search with an annoying interface.
Imagine if during the SaaS or big data or containerizaiton technology “revolutions” the application being run just didn’t matter at all. That’s kind of what’s going on with LLMs. Almost none of the products are all that much better than going to ChatGPT.com and dumping your data into the text box/file uploader and seeing what you get back.
Perhaps an analogy to describe what I mean would be if you were comparing two SaaS apps, like let’s say YNAB and the Simplifi budget app. In the world of the SaaS revolution, the capabilities of each application would be competitive advantages. I am choosing one over the other for the UX and feature list.
But in the AI LLM world, the difference between competing products is minimal. Whether you choose Cursor or Copilot or Firebase Studio you’re getting the same results because you’re feeding the same data to the same AI models. The companies that make the AI technologies basically don’t have a moat themselves, they’re basically just PaaS data center operators.
Your job? Might be secure for a lifetime, might be gone next week. No way to tell — "intelligence" isn't yet so well understood to just be an engineering challenge, but it is so well understood that the effect on jobs may be the same.
- https://www.clay.com/
- https://www.granola.ai/
There are a lot of tools in the sales space which fit your criteria.
Clay was founded 3 years before GPT 3 hit the market so I highly doubt that the majority of their core product runs on LLM-based AI. It is probably built on traditional machine learning.
When I ask it to spit out Svelte config files or something like that, I end up having to read the docs myself anyway because it can’t be trusted, for instance it will spew out tons of lines to configure every parameter as something that looks like the default when all it needs to do is follow the documentation that just uses defaults()
And it goes out of its way to “optimise” things that actually picks the wrong options versus the defaults which are fine.
The number of times I've had 2 reminders fire back-to-back because I asked Siri again to create one since I was _sure_ it didn't create the first one.
Siri is so dumb and it's insane that more heads have not rolled at Apple because of it (I'm aware of the recent shakeup, it's about a decade too late). Lastly, whoever decided to ship the new Siri UI without any of the new features should lose their job. What a squandered opportunity and effectively fraud IMHO.
More and more it's clear that Tim Cook is not the person that Apple needs at the helm. My mom knows Siri sucks, why doesn't the CEO and/or why is he incapable of doing anything to fix it. Get off your Trump-kissing, over-relying-on-China ass and fix your software! (Siri is not the only thing rotten)
Google is great at things like "Top 10 best rated movies of 2024", because people make lists of that sort of thing obsessively.
But Google is far less good at queries like "Which movies look visually beautiful but have been critically panned?". For that sort of thing I have far more luck with chatgpt because it's much less of a standard "top 10" list.
Interestingly, I despise that feature. It breaks the flow of what is actually a very simple task. Now I'm reading, reconsidering if the offered thing is the same thing I wanted over and over again.
The fact that I know this and spend time repeatedly disabling the damned things is awfully tiresome (but my fault for not paying for my own email etc etc)
Companies see that AI is a buzzword that means your stock goes up. So they start looking at it as an answer to the question: "How can I make my stock go up?" instead of "How can I create a better product", and then let the stock go up from creating a better product.
Of course, there's the upfront cost of Apple hardware... and the lack of server hardware per se... and Apple's seeming jekyll/hyde treatment of any use-case of their GPU's that doesn't involve their own direct business...
Or should there be an mega AI which will be my clone and can handle all these disparate scenarios in a unified manner?
Which approach will win ?
I feel the same though, AI allows me to debug stacktraces even quicker, because it can crunch through years of data on similar stack traces.
It is also a decent scaffolding tool, and can help fill in gaps when documentation is sparse, though its not always perfect.
Despite that, you also have tools like Apple Intelligence marketing the same thing, which are less dictated by metrics, in addition to doing it even less well.
It's layering AI into an existing workflow (and often saving a bit of time) but when you pull on the thread you fine more and more reasons that the workflow just shouldn't exist.
i.e. department A gets documents from department C, and they key them into a spreadsheet for department B. Sure LLMs can plug in here and save some time. But more broadly, it seems like this process shouldn't exist in the first place.
IMO this is where the "AI native" companies are going to just win out. It's not using AI as a bandaid over bad processes, but instead building a company in a way that those processes were never created in the first place.
I would bet AI-native companies acquire their own cruft over time.
A startup like Brex has a huge leg up on traditional banks when it comes to operational efficiency. And 99% of that is pre-ai. Just making online banking a first class experience.
But they've probably also built up a ton of cruft that some brand new startup won't.
(This is based on my knowledge the internal workings of a few well known tech companies.)
In my view there is significantly more there there with generative AI. But there is a huge amount of nonsense hype in both cases. So it has been fascinating to witness people in one case flailing around to find the meat on the bones while almost entirely coming up blank, while in the other case progressing on these parallel tracks where some people are mostly just responding to the hype while others are (more quietly) doing actual useful things.
To be clear, there was a period where I thought I saw a glimmer of people being on the "actual useful things" track in the blockchain world as well, and I think there have been lots of people working on that in totally good faith, but to me it just seems to be almost entirely a bust and likely to remain that way.
Meta is a behemoth. Google Plus, a footnote. The goal is to be Meta here and not Google Plus.
It baffles me how badly massive companies like Microsoft, Google, Apple etc are integrating AI into their products. I was excited about Gemini in Google sheets until I played around with it and realized it was barely usable (it specifically can’t do pivot tables for some reason? that was the first thing I tried it with lol).
This is a very fortunate truism for the kinds of builders and entrepreneurs who frequent this site! :)
To continue bashing on gmail/gemini, the worst offender in my opinion is the giant "Summarize this email" button, sitting on top of a one-liner email like "Got it, thanks". How much more can you possibly summarize that email?
https://koomen.dev/essays/horseless-carriages/#system-prompt...
MDX & claude are remarkably useful for expressing ideas. You could turn this into a little web app and it would instantly be better than any word processor ever created.
Here's the code btw https://github.com/koomen/koomen.dev
I'm with the vibe of wanting to move on to the point where LLMs are just yet another tool in the process of software engineering, and not the main focus.
More seriously, I think the reason companies don't want to expose the system prompt is because they want to keep some of the magic alive. Once most people understand that the universal interface to AI is text prompts, then all that will remain is the models themselves.
Go rewatch "The Forbin Project" from 1970.[1] Start at 31 minutes and watch to 35 minutes.
[1] https://archive.org/details/colossus-the-forbin-project-1970
WiFi?
We therefore connected Serif, which automatically writes drafts. You don't need to ask - open Gmail and drafts are there. Serif learned from previous support email threads to draft a proper response. And the tone matches!
I truly wonder why Gmail didn't think of that. Seems pretty obvious to me.
The interesting thing to think about is: Why are big mass audience products incentivized to ship more conservative and usually underwhelming implementations of new technology?
And then: What does that mean for the opportunity space for new products?
1. A new UX/UI paradigm. Writing prompts is dumb, re-writing prompts is even dumber. Chat interfaces suck.
2. "Magic" in the same way that Google felt like magic 25 years ago: a widget/app/thing that knows what you want to do before even you know what you want to do.
3. Learned behavior. It's ironic how even something like ChatGPT (it has hundreds of chats with me) barely knows anything about me & I constantly need to remind it of things.
4. Smart tool invocation. It's obvious that LLMs suck at logic/data/number crunching, but we have plenty of tools (like calculators or wikis) that don't. The fact that tool invocation is still in its infancy is a mistake. It should be at the forefront of every AI product.
5. Finally, we need PRODUCTS, not FEATURES; and this is exactly Pete's point. We need things that re-invent what it means to use AI in your product, not weirdly tacked-on features. Who's going to be the first team that builds an AI-powered operating system from scratch?
I'm working on this (and I'm sure many other people are as well). Last year, I worked on an MVP called Descartes[1][2] which was a spotlight-like OS widget. I'm re-working it this year after I had some friends and family test it out (and iterating on the idea of ditching the chat interface).
[1] https://vimeo.com/931907811
[2] https://dvt.name/wp-content/uploads/2024/04/image-11.png
I've wondered about this. Perhaps the concern is saved data will eventually overwhelm the context window? And so you must judicious in the "background knowledge" about yourself that gets remembered, and this problem is harder than it seems?
Btw, you can ask ChatGPT to "remember this". Ime the feature feels like it doesn't always work, but don't quote me on that.
https://youtu.be/ZUZT4x-detM
My thought is: Could the tool-routing layer be a much simpler "old school" NLP model? Then it would never try to do math and end up doing it poorly, because it just doesn't know how to do that. But you could give it a calculator tool and teach it how to pass queries along to that tool. And you could also give it a "send this to a people LLM tool" for anything that doesn't have another more targeted tool registered.
Is anyone doing it this way?
I'm working on a way of invoking tools mid-tokenizer-stream, which is kind of cool. So for example, the LLM says something like (simplified example) "(lots of thinking)... 1+2=" and then there's a parser (maybe regex, maybe LR, maybe LL(1), etc.) that sees that this is a "math-y thing" and automagically goes to the CALC tool which calculates "3", sticks it in the stream, so the current head is "(lots of thinking)... 1+2=3 " and then the LLM can continue with its thought process.
Look, I dunno if this idea makes sense, it's why I posed it as a question rather than a conviction. But I broadly have a sense that when a new technology hits, people are like "let's use it for everything!", and then as it matures, people find more success in interesting it with current approaches, or even trying older ideas but within the context of the new technology.
And it just strikes me that this "routing to tools" thing looks a lot like the part of expert systems that did work pretty well. But now we have the capability to make those tools themselves significantly smarter.
The problem is that AI is very often a way of hyping software. "This is a smart product. It is intelligent". It implies lightning in a bottle, a silver bullet. A new things that solves all your problems. But that is never true.
To create useful new stuff, to innovate, in a word, we need domain expertise and a lot of work. The world is full of complex systems and there are no short cuts. Well, there are, but there is always a trade off. You can pass it on (externalities) or you can hide (dishonesty) or you can use a sleight of hand and pretend the upside is so good, it's magical so just don't think about what it costs, ok? But it always costs something.
The promise of "expert systems" back then was creating "AI". It didn't happen. And there was an "AI winter" because people wised up to that shtick.
But then "big data" and "machine learning" collided in a big way. Transformers, "attention is all you need" and then ChatGPT. People got this warm fuzzy feeling inside. These chatbots got impressive, and improved fast! It was quite amazing. It got A LOT of attention and has been driving a lot of investment. It's everywhere now, but it's becoming clear it is falling very short of "AI" once again. The promised land turned out once again to just be someone else's land.
So when people look at this attempt at AI and its limitations, and start wondering "hey what if we did X" and X sounds just like what people were trying when we last thought AI might just be around the corner... Well let's just say I am having a deja vu.
It's fine to have a hobby horse! I certainly have lots of them!
But I'm sorry, it's just not relevant to this thread.
Edit to add: To be clear, it may very well be a good point! It's just not what I was talking about here.
> I think it's an expert system
I respectfully disagree with the claim that my point is petty and irrelevant in this context.
My question at the beginning of the thread was: Assuming people are using a particular pattern, where LLMs are used to parse prompts and route them to purpose-specific tools (which is what the thread I was replying in is about), is it actually a good use of LLMs to implement that routing layer, or mightn't we use a simpler implementation for the routing layer?
Your point seems more akin to questioning whether the entire concept of farming out to tools makes sense. Which is interesting, but just a different discussion.
> I didn't say it's petty!
You did.
And I already showed you made a claim that LLM was AI and that you agree that you were thinking of something akin to expert systems. When I explained why I think this is a signal that we are headed to another AI winter you started deflecting.
I am done with this conversation.
I don't think I've deflected at all. You're just talking about something orthogonal to the thing I was asking about in this thread, which (predictably) totally derailed the thread into an entirely different discussion.
It's fine! Happens all the time.
E.g. Scott Aaronson | How Much Math Is Knowable?
https://youtu.be/VplMHWSZf5c
The video slides could be converted into a dark mode for night viewing.
> 2. "Magic" in the same way that Google felt like magic 25 years ago: a widget/app/thing that knows what you want to do before even you know what you want to do.
and not to "dunk" on you or anything of the sort but that's literally what Descartes seems to be? Another wrapper where I am writing prompts telling the AI what to do.
Not at all, you're totally correct; I'm re-imagining it this year from scratch, it was just a little experiment I was working on (trying to combine OS + AI). Though, to be clear, it's built in rust & it fully runs models locally, so it's not really a ChatGPT wrapper in the "I'm just calling an API" sense.
You might be interested in this series: https://www.youtube.com/@liber-indigo
In the same way that Microsoft and the 'IBM clones' brought us the current computing paradigm built on the desktop metaphor, I believe there will have to be a new OS built on a new metaphor. It's just a question of when those perfect conditions arise for lightning to strike on the founders who can make it happen. And just like Xerox and IBM, the actual core ideas might come from the tech giants (FAANG et al.) but they may not end up being the ones to successfully transition to the new modality.
A feature that seems to me would truly be "smart" would be an e-mail client that observes my behavior over time and learns from it directly. Without me prompting or specifying rules at all, it understands and mimics my actions and starts to eventually do some of them automatically. I suspect doing that requires true online learning, though, as in the model itself changes over time, rather than just adding to a pre-built prompt injected to the front of a context window.
> Does this mean I always want to write my own System Prompt from scratch? No. I've been using Gmail for twenty years; Gemini should be able to write a draft prompt for me using my emails as reference examples.
This is where it'll get hard for teams who integrate AI into things. Not only is retrieval across a large set of data hard, but this also implies a level of domain expertise on how to act that a product can help users be more successful with. For example, if the product involves data analysis, what are generally good ways to actually analyze the data given the tools at hand? The end-user often doesn't know this, so there's an opportunity to empower them ... but also an opportunity to screw it up and make too many assumptions about what they actually want to do.
But for the 99 other messages, especially things that mundanely convey information like "My daughter has the flu and I won't be in today", "Yes 2pm at Shake Shack sounds good", it will be much faster to read over drafts that are correct and then click send.
The only reason this wouldn't be faster is if the drafts are bad. And that is the point of the article: the models are good enough now that AI drafts don't need to be bad. We are just used to AI drafts being bad due to poor design.
Do you really run these things through an AI to burden your reader with pointless additional text?
It’s like you’re asking why you would want a password manager when you can just type the characters yourself. It saves time if done correctly.
Most of us don't need to write the CEO email ever in our life. I assume the CEO will write the flu message to his staff in the same style of tone as everyone else.
For contrast:
"All: my daughter is home sick, I won't be in the office today" (CEO style)
vs
"Hi everyone, I'm very sorry to make this change last minute but due to an unexpected illness in the family, I'll need to work from home today and won't be in the office at my usual time. My daughter has the flu and could not go to school. Please let me know if there are any questions, I'll be available on Slack if you need me." (not CEO style)
An AI summary of the second message might look something like the first message.
I know what you are trying to say. I agree that for most emails that first tone is better. However when you need to send something to a large audience the second is better.
But it's handy when the recipient is less familiar. When I'm writing to my kid's school's principal about some issue, I can't really say, "Susan's lunch money got stolen. Please address it." There has to be more. And it can be hard knowing what that needs to be, especially for a non-native speaker. LLMs tend to take it too far in the other direction, but you can get it to tone it down, or just take the pieces that you like.
Why?
I mean this sincerely. Why is the message you quoted not enough?
I don't always _feel_ autistic, but stuff like this reminds me that I'm not normal.
Being too flowery and indirect is annoying but not impolite. If you overdo it then people may still get annoyed with you, but for different reasons. For most situations you don’t need too much, a salutation and a “I hope you’re doing well” and a brief mention of who you are and what you’re writing about can suffice.
And we’re talking micro optimisation here.
I mean I’ve sent 23 emails this year. Yeah that’s it.
It takes me all of 5 seconds to type messages like that (I timed myself typing it). Where exactly is the savings from AI? I don't care, at all, if a 5s process can be turned into a 2s process (which I doubt it even can).
However, I do know people who are not native speakers, or who didn't do an advanced degree that required a lot of writing, and they report loving the ability to have it clean up their writing in professional settings.
This is fairly niche, and already had products targeting it, but it is at least one useful thing.
My email inbox is already filled with a bunch of automated emails that provide me no info and waste my time. The last thing I want is an AI tool that makes it easier to generate even more crap.
Emails sent company-wide need to be especially short, because so many person-hours are spent reading them. Also, they need to provide the most background context to be understood, because most of those readers won't already share the common ground to understand a compressed message, increasing the risk of miscommunication.
This is why messages need to be extremely brief, but also not.
Edit: added /s because it wasn't apparent this was sarcastic
https://news.ycombinator.com/item?id=42712143
How is AI in email a good thing?!
There's a cartoon going around where in the first frame, one character points to their screen and says to another: "AI turns this single bullet point list into a long email I can pretend I wrote".
And in the other frame, there are two different characters, one of them presumably the receiver of the email sent in the first frame, who says to their colleague: "AI makes a single bullet point out of this long email I can pretend I read".
The cartoon itself is the one posted above by PyWoody.
In other contexts, someone I knew had written a system to generate automated emails in response to various online events. They later ran into someone who'd written automated processing systems to act on those emails. This made the original automater quite happy.
(Context crossed organisational / institutional boundaries, there was no explicit coordination between the two.)
And I have to wonder, why? What's the point?
This point is made multiple times in the article (which is very good; I recommend reading it!):
> The email I'd have written is actually shorter than the original prompt, which means I spent more time asking Gemini for help than I would have if I'd just written the draft myself. Remarkably, the Gmail team has shipped a product that perfectly captures the experience of managing an underperforming employee.
> As I mentioned above, however, a better System Prompt still won't save me much time on writing emails from scratch. The reason, of course, is that I prefer my emails to be as short as possible, which means any email written in my voice will be roughly the same length as the User Prompt that describes it. I've had a similar experience every time I've tried to use an LLM to write something. Surprisingly, generative AI models are not actually that useful for generating text.
the important point of communicating is to get the other person to understand you. if my own words fall flat for whatever reason, if there are better words to use, I'd prefer to use those instead.
"fuck you, pay me" isn't professional communication with a client. a differently worded message might be more effective (or not). spending an hour agonizing over what to say is easier spent when you have someone help you write it
I could have written them a message saying "Zendesk has views, not folders [and figure out what I mean by that]", but instead I asked AI something like:
The clear, detailed explanation I got was useful for my colleague, and required little effort from me (after the initial diagnosis).You could even skip the custom system prompt entirely and just have it analyze a randomized but statistically-significant portion of the corpus of your outgoing emails and their style, and have it replicate that in drafts.
You wouldn't even need a UI for this! You could sell a service that you simply authenticated to your inbox and it could do all this from the backend.
It would likely end up being close enough to the mark that the uncanny valley might get skipped and you would mostly just be approving emails after reviewing them.
Similar to reviewing AI-generated code.
The question is, is this what we want? I've already caught myself asking ChatGPT to counterargue as me (but with less inflammatory wording) and it's done an excellent job which I've then (more or less) copy-pasted into social-media responses. That's just one step away from having them automatically appear, just waiting for my approval to post.
Is AI just turning everyone into a "work reviewer" instead of a "work doer"?
A lot of work is inherently repetitive, or involves critical but burdensome details. I'm not going to manually write dozens of lines of code when I can do `bin/rails generate scaffold User name:string`, or manually convert decimal to binary when I can access a calculator within half a second. All the important labor is in writing the prompt, reviewing the output, and altering it as desired. The act of generating the boilerplate itself is busywork. Using a LLM instead of a fixed-functionality wizard doesn't change this.
The new thing is that the generator is essentially unbounded and silently degrades when you go beyond its limits. If you want to learn how to use AI, you have to learn when not to use it.
Using AI for social media is distinct from this. Arguing with random people on the internet has never been a good idea and has always been a massive waste of time. Automating it with AI just makes this more obvious. The only way to have a proper discussion is going to be face-to-face, I'm afraid.
This isn't the tautology you imagine it to be.
Consider the example given here of having AI write one line draft response to emails. To validate such response, you have to: (1) read the original email, (2) understand it, (3) decide what you want to communicate in your reply, then (4) validate that the suggested draft communicates the same.
If the AI gave a correct answer, you saved yourself from typing one sentence, which you probably already formulated in your head in step (3). A minor help, at best.
But if the AI was wrong, you now have to write that reply yourself.
To get positive expected utility from the above scenario, you'd need the probability of the AI to be correct extremely high, and even then, the savings would be small.
A task that requires more effort to turn ideas into deliverables would have better expectation, but complex tasks often have results that are not simple nor easy to check, so the savings may not be as meaningful as you naively assume.
to address your larger point, I think AI-generated drafts written in my voice will be helpful for mundane, transaction emails, but not for important messages. Even simple questions like "what do you feel like doing for dinner tonight" could only be answered by me, and that's fine. If an AI can manage my inbox while I focus on the handful of messages that really need my time and attention that would be a huge win in my book.
Just a reminder that these things still need significant oversight or very targeted applications, I suppose.
The email labeling assistant is a great example of this. Most mail services can already do most of this, so the best-case scenario is using AI to translate your human speech into a suggestion for whatever format the service's rules engine uses. Very helpful, not flashy: you set it up once and forget about it.
Being able to automatically interpret the "Reschedule" email and suggest a diff for an event in your calendar is extremely useful, as it'd reduce it to a single click - but it won't be flashy. Ideally you wouldn't even notice there's a LLM behind it, there's just a "confirm reschedule button" which magically appears next to the email when appropriate.
Automatically archiving sales offers? That's a spam filter. A really good one, mind you, but hardly something to put on the frontpage of today's newsletters.
It can all provide quite a bit of value, but it's simply not sexy enough! You can't add a flashy wizard staff & sparkles icon to it and charge $20 / month for that. In practice you might be getting a car, but it's going to look like a horseless carriage to the average user. They want Magic Wizard Stuff, not invest hours into learning prompt programming.
I don't have a lot of doubt that it is technically doable, but it's not going to be economically viable when it has to pay back hundreds of billions of dollars of investments into training models and buying shiny hardware. The industry first needs to get rid of that burden, which means writing off the training costs and running inference on heavily-discounted supernumerary hardware.
I'll believe this when I stop spending so much time deleting email I don't want to read.
i like that :)
In my experience there is a vague divide between the things that can and can't be created using LLMs. There's a lot of things where AI is absolutely a speed boost. But from a certain point, not so much, and it can start being an impediment by sending you down wrong paths, and introducing subtle bugs to your code.
I feel like the speedup is in "things that are small and done frequently". For example "write merge sort in C". Fast and easy. Or "write a Typescript function that checks if a value is a JSON object and makes the type system aware of this". It works.
"Let's build a chrome extension that enables navigating webpages using key chords. it should include a functionality where a selected text is passed to an llm through predefined prompts, and a way to manage these prompts and bind them to the chords." gives us some code that we can salvage, but it's far from a complete solution.
For unusual algorithmic problems, I'm typically out of luck.
Sounded like a cool idea on first read, but when thinking how to apply personally, I can't think of a single thing I'd want to set up autoreply for, even drafts. Email is mostly all notifications or junk. It's not really two-way communication anymore. And chat, due to its short form, doesn't benefit much from AI draft.
So I don't disagree with the post, but am having trouble figuring out what a valid use case would be.
The simple answer is that they lose their revenue if you aren’t actually reading the emails. The reason you need this feature in the first place is because you are bombarded with emails that don’t add any value to you 99% of the time. I mean who gets that many emails really? The emails that do get to you get Google some money in exchange for your attention. If at any point it’s the AI that’s reading your emails, Google suddenly cannot charge money they do now. There will be a day when they ship this feature, but that will be a day when they figure out how to charge money to let AI bubble up info that makes them money, just like they did it in search.
Clearly that's nonsense. They want you to use Gmail because they want you to stay in the Google ecosystem and if you switch to a competitor they won't get any money at all. The reason they don't have AI to categorise your emails is that LLMs that can do it are extremely new and still relatively unreliable. It will happen. In fact it already did happen with Inbox, and I think normal gmail had promotion filtering for a while.
It’s the same reason you see an ad on Facebook after every couple of posts. But you will neither see a constant stream of ads nor a completely ad free experience.
AKA make it look that the email reply was not written by an AI
> I'm a GP at YC
So you are basically out-sourcing your core competence to AI. You could just skip a step and set up an auto-reply like "please ask Gemini 2.5 what an YC GP would reply to your request and act accordingly"