7 ms·

This is easily among the rare highest quality articles/comments I've read in the past weeks, perhaps months (on LLMs/AI since that's what I am particularly inte

This is easily among the rare highest quality articles/comments I've read in the past weeks, perhaps months (on LLMs/AI since that's what I am particularly interested in). And this was for internal consumption before it was made public. Reinforces my recent impression that so much that's being made for public consumption now is shallow and it is hard to find the good stuff. And sadly, increasing so even on HN. As I write this, I acknowledge I discovered this on HN :) Wish we had ways to incentivize the public sharing of such high-quality content that don't die at the altar of micro rewards.

censor_me 3y ago
[dead]
visarga 3y ago
I've been saying the same things for weeks, right here and in the usual places. Basically - OpenAI will not be able to continue to commercialise chatGPT-3.5, they will have to move to GPT-4 because the open source alternatives will catch up. Their island of exclusivity is shrinking fast. In a few months nobody will want to pay for GPT-4 either when they can have private, cheap equivalents. So GPT-5 it is for OpenAI. But the bulk of the tasks can probably be solved at 3.5 level, another more difficult chunk with 4, I'm wondering how many of the requests will be so complex as to require GPT-5. Probably less than 1%. There's a significant distinction between web search and generative AI. You can't download "a Google" but you can download "a LLaMA". This marks the end of the centralisation era and increased user freedom. Engaging in chat and image generation without being tracked is now possible while searching, browsing the web or torrenting are still tracked.
- com2kid 3y ago
  > I've been saying the same things for weeks, right here and in the usual places. Basically - OpenAI will not be able to continue to commercialise chatGPT-3.5, they will have to move to GPT-4 because the open source alternatives will catch up. Their island of exclusivity is shrinking fast. In a few months nobody will want to pay for GPT-4 either when they can have private, cheap equivalents. So GPT-5 it is for OpenAI. It is worth $20 a month to have one UI on one service that does everything. Unless specialized models can far exceed what GPT4 can do, being general purpose is amazing. IMHO the future is APIs written for consumption by LLMs, and then natural language interfaces and just telling an AI literally anything you want done.
  - Barrin92 3y ago
    >It is worth $20 a month to have one UI on one service that does everything. competition will drive profit margins and prices down to nothing because the number of companies that can spin up an UI is unlimited. Markets don't pay you what something is worth, they pay what the cheapest participant is willing to sell it for.
    - com2kid 3y ago
      > competition will drive profit margins and prices down to nothing I strongly suspect the profit margin on ChatGPT is already pretty low! > Markets don't pay you what something is worth, they pay what the cheapest participant is willing to sell it for. Correction: Markets pay what companies are able to convince consumers to pay. Some products bring negative value to the buyer, but are still sold for hundreds of millions of dollars (see: enterprise sales and integrations, which oftentime fail).
    - bryanrasmussen 3y ago
      >Markets don't pay you what something is worth, they pay what the cheapest participant is willing to sell it for. I believe 'what something is worth' is defined as what the market is willing to pay. And sometimes the customer will pay for something that isn't the cheapest of something, which is why I'm writing this on a mac.
    - birdyrooster 3y ago
      That last argument is a tautology btw
  - xiphias2 3y ago
    I'm paying but hate the UI. I had to add labels myself as a Tampermonkey extension, but it would be much better if they would give API access to what I'm paying for and let UIs compete.
  - dragonwriter 3y ago
    > It is worth $20 a month to have one UI on one service that does everything. Today it is. When there is an open source, capable “one UI for everything” that runs locally and can consume external services as needed (but keeps your data locally otherwise), will it still be?
- deanc 3y ago
  And where are these open source models where I can go to a url and do all the things I can do in ChatGPT or through api keys for OpenAI? I googled a couple of weeks ago to find hosted versions of these open source models to try, and every one was either down or woefully poor. OpenAI and MS are going to win because they have a package to go and it’s ready and available and working well - they have set the benchmark. I’m not seeing any evidence of this in the OSS community thus far. Until I can spin up a docker image capable of the same as OpenAI in hetzner for 30 bucks a month - it’s not in the same league.
  - icyfox 3y ago
    One issue with the current generation of open source models is most have been based on some llama core architecture, and that's not licensed for commercial use. Once you get to the point of spinning up a full and easy API, and selling API credentials, you're entering into the commercial clause. Once we have a llama alternative (or a more permissively licensed separate architecture) I guarantee hosting providers like Render or Model are going to come in with an API offering. Just waiting on those core models to improve licensing, would be my guess.
  - dragonwriter 3y ago
    > Until I can spin up a docker image capable of the same as OpenAI in hetzner for 30 bucks a month - it’s not in the same league. Yes, you are right That’s irrelevant to the point of this, which is about the dynamics of the market over a longer window than “what is available to use immediately today”, because a “moat” is a different thing than “a current lead”.
  - MacsHeadroom 3y ago
    >Until I can spin up a docker image capable of the same as OpenAI in hetzner for 30 bucks a month I do exactly this with https://github.com/nsarrazin/serge https://github.com/nsarrazin/serge Hetzner will install any hardware you send them for $100. So you can send them a $200 P40 24GB to run 33B parameter GPU models at ChatGPT speeds without increasing your monthly cost.
    - 3y ago
- huijzer 3y ago
  I also would like to believe that, but there are countless examples which show the difference. Companies have no time to figure out which of the open source offerings is the best. Even worse, they don’t have the time to switch from one project to the other or back to OpenAI if OpenAI releases a new state-of-the-art model.
- digging 3y ago
  > I'm wondering how many of the requests will be so complex as to require GPT-5 I am not sure the pessimism is warranted. True that few people have the need to upgrade from GPT-3.5 to GPT-4 now, but if GPT-5 is another serious leap in capabilities, it might have an effect closer to the difference between old chatbots (useless, interesting) and ChatGPT (immediate economic impact, transforming some jobs). Or at any rate, we should expect such a leap to occur soon, even if it's not GPT-5.
  - ngngngng 3y ago
    Also significant to note that much of this AI boom was due to the UI of ChatGPT that gave everyone easy access to the model. Perhaps much of the improvements to be had in GPT-5 will also be found in the UI. I mean UI in the broadest possible sense, I'm sure we'll come up with very creative ways to interact with this over the coming years. But the moat problem addressed in the article remains. Good luck patenting your amazing UI change in such a way that open source models can't catch up within a few weeks.
- tlonny 3y ago
  > I've been saying the same things for weeks, right here and in the usual places. Basically - OpenAI will not be able to continue to commercialise chatGPT-3.5, they will have to move to GPT-4 because the open source alternatives will catch up. Their island of exclusivity is shrinking fast. In a few months nobody will want to pay for GPT-4 either when they can have private, cheap equivalents. So GPT-5 it is for OpenAI. I wonder if this effect will be compounded by regulatory pressure that seems poised to slow down progress at the bleeding edge of LLMs. Open source closing the gap at the bottom, and governments restricting further movement at the top...
whimsicalism 3y ago
If you feel like your criteria for quality is beyond what you can typically find in the popular public consumption, just start reading papers directly?
- crazygringo 3y ago
  The value in this article is the business strategy perspective, not the details of LLM's. You generally won't find papers detailing the present-moment business strategies of specific for-profit corporations.
  - whimsicalism 3y ago
    Sure but this article is also not the present-moment business strategy, it is written by a single individual with a perspective.
crazygringo 3y ago
Well yes, generally in the business world all the "good stuff", the really smart analysis, is extremely confidential. Really smart people are putting these things together, but these types of analyses are a competitive advantage, so they're absolutely never going to share it publicly. This was leaked, not intentionally made public. And it all makes sense -- the people producing these types of business analyses are world-class experts in their fields (the business strategy not just the tech), and are paid handsomely for that. The "regular stuff" people consume is written by journalists who are usually a bit more "jack of all trades master of none". A journalist might cover the entire consumer tech industry, not LLM's specifically. They can't produce this kind of analysis, nor should we expect them to. Industry experts are extremely valuable for a reason, and they don't bother writing analyses for public media since it doesn't pay as well.
- wahern 3y ago
  Sort of like sports recruiters.
- gerad 3y ago
  Beware that there’s also a ton of bias when something is analyzed internally. As Upton Sinclair once said “It is difficult to get a man to understand something, when his salary depends on his not understanding it.” In the case of this analysis - it sounds great but it’s wrong. OpenAI has a huge moat. It has captured the mind share of the world. The software it has shipped is dramatically better than anything anyone else has shipped (the difference between useless and useful). We’ll see if folks catch up, but the race is currently OpenAI’s to lose.
  - crazygringo 3y ago
    Mind share is not a moat. And market share or being first is not a moat. Moats are very specific things such as network effects, customers with sunk costs, etc. The very point of the term "moat" is to distinguish it from things like market share or mind share. The article is correct, OpenAI has no moat currently.
    - 3y ago
burnished 3y ago
Most of it is being written to make money off of you instead of communicate with you and it shows.
seydor 3y ago
a lot of people have said similar things here
jiggywiggy 3y ago
Im a noob. But the time for Wikipedia language models & training models seems ripe.
- PeterCorless 3y ago
  • https://en.wikipedia.org/wiki/Large_language_model#List_of_large_language_models https://en.wikipedia.org/wiki/Large_language_model#List_of_l... • https://en.wikipedia.org/wiki/List_of_datasets_for_machine-learning_research#Internet https://en.wikipedia.org/wiki/List_of_datasets_for_machine-l...
swores 3y ago
Hi Sai, do you have an email address (or other preferred private message) I could contact you on? Feel free to send it to the relay email in my profile if you want to avoid putting it publicly (or reply here how to contact you). I'll ask my first question here below, so that if you have an answer it can benefit other HNers, and I'll save the other line of thought for email. Do you happen to have a list of other highest quality articles on AI/LLMs/etc that you've come across, and could share here? It's not my field but something I want to learn more about, and I've found it hard to, without knowing much about the specific subjects within AI that would be good to learn about makes it hard picking what to read or not.
0xbadcafebee 3y ago
Most HN submissions are clickbait advertisements by startups for B2B/B2C services, clickbait amateur blog editorials looking for subscribers, tutorials for newbies, conspiracy theories, spam, and literally every article posted to a major media outlet. Most comments are by amateurs that sound really confident. Don't believe me? Go look at https://news.ycombinator.com/newest https://news.ycombinator.com/newest . Maybe once a month you find something on here that is actually from an expert who knows what they're talking about and hasn't written a book on it yet, or a pet project by an incredibly talented person who has no idea it was submitted. Source: I've been here for 14 years. That makes me a little depressed...
opportune 3y ago
There is some really high quality internal discussions at tech companies, unfortunately they are suffering from leaks due to their size and media have realized it’s really easy to just take their internal content and publish it. It really sucks because there’s definitely a chilling effect knowing any personal opinion expressed in text at a big tech company could end up in a headline like “GOOGLE SAYS <hot take>” because of a leak. If there is some kind of really bad behavior being exposed, I think the role of the media is to help do that. But I don’t think their role should be to expose any leaked internal document they can get their hands on.
- UncleMeat 3y ago
  This is exactly that. This doc is apparently a leaked internal doc.
  - opportune 3y ago
    I know that, my point is that it’s not indicating anything nefarious enough to be worth exposing, it’s just juicy. I don’t think the media should share stuff like this just because it’s interesting. They’re making a market for corporate espionage to sell clicks.
heliophobicdude 3y ago
I thought this was a good one this week but didn't get popular. https://huyenchip.com/2023/05/02/rlhf.html https://huyenchip.com/2023/05/02/rlhf.html

censor_me 3y ago

[dead]

visarga 3y ago

I've been saying the same things for weeks, right here and in the usual places. Basically - OpenAI will not be able to continue to commercialise chatGPT-3.5, they will have to move to GPT-4 because the open source alternatives will catch up. Their island of exclusivity is shrinking fast. In a few months nobody will want to pay for GPT-4 either when they can have private, cheap equivalents. So GPT-5 it is for OpenAI. But the bulk of the tasks can probably be solved at 3.5 level, another more difficult chunk with 4, I'm wondering how many of the requests will be so complex as to require GPT-5. Probably less than 1%. There's a significant distinction between web search and generative AI. You can't download "a Google" but you can download "a LLaMA". This marks the end of the centralisation era and increased user freedom. Engaging in chat and image generation without being tracked is now possible while searching, browsing the web or torrenting are still tracked.

com2kid 3y ago
> I've been saying the same things for weeks, right here and in the usual places. Basically - OpenAI will not be able to continue to commercialise chatGPT-3.5, they will have to move to GPT-4 because the open source alternatives will catch up. Their island of exclusivity is shrinking fast. In a few months nobody will want to pay for GPT-4 either when they can have private, cheap equivalents. So GPT-5 it is for OpenAI. It is worth $20 a month to have one UI on one service that does everything. Unless specialized models can far exceed what GPT4 can do, being general purpose is amazing. IMHO the future is APIs written for consumption by LLMs, and then natural language interfaces and just telling an AI literally anything you want done.
- Barrin92 3y ago
  >It is worth $20 a month to have one UI on one service that does everything. competition will drive profit margins and prices down to nothing because the number of companies that can spin up an UI is unlimited. Markets don't pay you what something is worth, they pay what the cheapest participant is willing to sell it for.
  - com2kid 3y ago
    > competition will drive profit margins and prices down to nothing I strongly suspect the profit margin on ChatGPT is already pretty low! > Markets don't pay you what something is worth, they pay what the cheapest participant is willing to sell it for. Correction: Markets pay what companies are able to convince consumers to pay. Some products bring negative value to the buyer, but are still sold for hundreds of millions of dollars (see: enterprise sales and integrations, which oftentime fail).
  - bryanrasmussen 3y ago
    >Markets don't pay you what something is worth, they pay what the cheapest participant is willing to sell it for. I believe 'what something is worth' is defined as what the market is willing to pay. And sometimes the customer will pay for something that isn't the cheapest of something, which is why I'm writing this on a mac.
  - birdyrooster 3y ago
    That last argument is a tautology btw
- xiphias2 3y ago
  I'm paying but hate the UI. I had to add labels myself as a Tampermonkey extension, but it would be much better if they would give API access to what I'm paying for and let UIs compete.
- dragonwriter 3y ago
  > It is worth $20 a month to have one UI on one service that does everything. Today it is. When there is an open source, capable “one UI for everything” that runs locally and can consume external services as needed (but keeps your data locally otherwise), will it still be?
deanc 3y ago
And where are these open source models where I can go to a url and do all the things I can do in ChatGPT or through api keys for OpenAI? I googled a couple of weeks ago to find hosted versions of these open source models to try, and every one was either down or woefully poor. OpenAI and MS are going to win because they have a package to go and it’s ready and available and working well - they have set the benchmark. I’m not seeing any evidence of this in the OSS community thus far. Until I can spin up a docker image capable of the same as OpenAI in hetzner for 30 bucks a month - it’s not in the same league.
- icyfox 3y ago
  One issue with the current generation of open source models is most have been based on some llama core architecture, and that's not licensed for commercial use. Once you get to the point of spinning up a full and easy API, and selling API credentials, you're entering into the commercial clause. Once we have a llama alternative (or a more permissively licensed separate architecture) I guarantee hosting providers like Render or Model are going to come in with an API offering. Just waiting on those core models to improve licensing, would be my guess.
- dragonwriter 3y ago
  > Until I can spin up a docker image capable of the same as OpenAI in hetzner for 30 bucks a month - it’s not in the same league. Yes, you are right That’s irrelevant to the point of this, which is about the dynamics of the market over a longer window than “what is available to use immediately today”, because a “moat” is a different thing than “a current lead”.
- MacsHeadroom 3y ago
  >Until I can spin up a docker image capable of the same as OpenAI in hetzner for 30 bucks a month I do exactly this with https://github.com/nsarrazin/serge https://github.com/nsarrazin/serge Hetzner will install any hardware you send them for $100. So you can send them a $200 P40 24GB to run 33B parameter GPU models at ChatGPT speeds without increasing your monthly cost.
  - 3y ago
huijzer 3y ago
I also would like to believe that, but there are countless examples which show the difference. Companies have no time to figure out which of the open source offerings is the best. Even worse, they don’t have the time to switch from one project to the other or back to OpenAI if OpenAI releases a new state-of-the-art model.
digging 3y ago
> I'm wondering how many of the requests will be so complex as to require GPT-5 I am not sure the pessimism is warranted. True that few people have the need to upgrade from GPT-3.5 to GPT-4 now, but if GPT-5 is another serious leap in capabilities, it might have an effect closer to the difference between old chatbots (useless, interesting) and ChatGPT (immediate economic impact, transforming some jobs). Or at any rate, we should expect such a leap to occur soon, even if it's not GPT-5.
- ngngngng 3y ago
  Also significant to note that much of this AI boom was due to the UI of ChatGPT that gave everyone easy access to the model. Perhaps much of the improvements to be had in GPT-5 will also be found in the UI. I mean UI in the broadest possible sense, I'm sure we'll come up with very creative ways to interact with this over the coming years. But the moat problem addressed in the article remains. Good luck patenting your amazing UI change in such a way that open source models can't catch up within a few weeks.
tlonny 3y ago
> I've been saying the same things for weeks, right here and in the usual places. Basically - OpenAI will not be able to continue to commercialise chatGPT-3.5, they will have to move to GPT-4 because the open source alternatives will catch up. Their island of exclusivity is shrinking fast. In a few months nobody will want to pay for GPT-4 either when they can have private, cheap equivalents. So GPT-5 it is for OpenAI. I wonder if this effect will be compounded by regulatory pressure that seems poised to slow down progress at the bleeding edge of LLMs. Open source closing the gap at the bottom, and governments restricting further movement at the top...

whimsicalism 3y ago

If you feel like your criteria for quality is beyond what you can typically find in the popular public consumption, just start reading papers directly?

crazygringo 3y ago
The value in this article is the business strategy perspective, not the details of LLM's. You generally won't find papers detailing the present-moment business strategies of specific for-profit corporations.
- whimsicalism 3y ago
  Sure but this article is also not the present-moment business strategy, it is written by a single individual with a perspective.

crazygringo 3y ago

Well yes, generally in the business world all the "good stuff", the really smart analysis, is extremely confidential. Really smart people are putting these things together, but these types of analyses are a competitive advantage, so they're absolutely never going to share it publicly. This was leaked, not intentionally made public. And it all makes sense -- the people producing these types of business analyses are world-class experts in their fields (the business strategy not just the tech), and are paid handsomely for that. The "regular stuff" people consume is written by journalists who are usually a bit more "jack of all trades master of none". A journalist might cover the entire consumer tech industry, not LLM's specifically. They can't produce this kind of analysis, nor should we expect them to. Industry experts are extremely valuable for a reason, and they don't bother writing analyses for public media since it doesn't pay as well.

wahern 3y ago
Sort of like sports recruiters.
gerad 3y ago
Beware that there’s also a ton of bias when something is analyzed internally. As Upton Sinclair once said “It is difficult to get a man to understand something, when his salary depends on his not understanding it.” In the case of this analysis - it sounds great but it’s wrong. OpenAI has a huge moat. It has captured the mind share of the world. The software it has shipped is dramatically better than anything anyone else has shipped (the difference between useless and useful). We’ll see if folks catch up, but the race is currently OpenAI’s to lose.
- crazygringo 3y ago
  Mind share is not a moat. And market share or being first is not a moat. Moats are very specific things such as network effects, customers with sunk costs, etc. The very point of the term "moat" is to distinguish it from things like market share or mind share. The article is correct, OpenAI has no moat currently.
  - 3y ago

burnished 3y ago

Most of it is being written to make money off of you instead of communicate with you and it shows.

seydor 3y ago

a lot of people have said similar things here

jiggywiggy 3y ago

Im a noob. But the time for Wikipedia language models & training models seems ripe.

PeterCorless 3y ago
• https://en.wikipedia.org/wiki/Large_language_model#List_of_large_language_models https://en.wikipedia.org/wiki/Large_language_model#List_of_l... • https://en.wikipedia.org/wiki/List_of_datasets_for_machine-learning_research#Internet https://en.wikipedia.org/wiki/List_of_datasets_for_machine-l...

swores 3y ago

Hi Sai, do you have an email address (or other preferred private message) I could contact you on? Feel free to send it to the relay email in my profile if you want to avoid putting it publicly (or reply here how to contact you). I'll ask my first question here below, so that if you have an answer it can benefit other HNers, and I'll save the other line of thought for email. Do you happen to have a list of other highest quality articles on AI/LLMs/etc that you've come across, and could share here? It's not my field but something I want to learn more about, and I've found it hard to, without knowing much about the specific subjects within AI that would be good to learn about makes it hard picking what to read or not.

0xbadcafebee 3y ago

Most HN submissions are clickbait advertisements by startups for B2B/B2C services, clickbait amateur blog editorials looking for subscribers, tutorials for newbies, conspiracy theories, spam, and literally every article posted to a major media outlet. Most comments are by amateurs that sound really confident. Don't believe me? Go look at https://news.ycombinator.com/newest https://news.ycombinator.com/newest . Maybe once a month you find something on here that is actually from an expert who knows what they're talking about and hasn't written a book on it yet, or a pet project by an incredibly talented person who has no idea it was submitted. Source: I've been here for 14 years. That makes me a little depressed...

opportune 3y ago

There is some really high quality internal discussions at tech companies, unfortunately they are suffering from leaks due to their size and media have realized it’s really easy to just take their internal content and publish it. It really sucks because there’s definitely a chilling effect knowing any personal opinion expressed in text at a big tech company could end up in a headline like “GOOGLE SAYS <hot take>” because of a leak. If there is some kind of really bad behavior being exposed, I think the role of the media is to help do that. But I don’t think their role should be to expose any leaked internal document they can get their hands on.

UncleMeat 3y ago
This is exactly that. This doc is apparently a leaked internal doc.
- opportune 3y ago
  I know that, my point is that it’s not indicating anything nefarious enough to be worth exposing, it’s just juicy. I don’t think the media should share stuff like this just because it’s interesting. They’re making a market for corporate espionage to sell clicks.

heliophobicdude 3y ago

I thought this was a good one this week but didn't get popular. https://huyenchip.com/2023/05/02/rlhf.html https://huyenchip.com/2023/05/02/rlhf.html