Is this the kind of thing employees had in mind when they backed Altman?
(FWIW, I ended up not taking the position, mainly because I would not have enjoyed the specific work they wanted to hire me for.)
I had a badly timed offer of something (I don't really remember might just have been an interview, or contracting work, on the back of some OSS contribution) at Cloudflare pre-IPO, and I do occasionally regret/wonder what-if, because it's not just a random hypothetical, it was an actual opportunity I made a decision against. (Doesn't help that it's still a company on my interested-in list, much like the trap in investing of liking a stock but wishing you'd bought it lower when you first thought of it, and so continuing to not buy as it climbs (or not even investing necessarily but just being a consumer in an inflationary environment).)
This seems very much like a "get out now" signal to everybody involved. Greater fool farming on an accelerated timeline.
Not even Google can compete.
It could also be a signal that the company recognizes that they would not be where they currently are without their workers. Maybe they're giving a reason for their workers to stay (become multi-millionaires overnight) and not go to their competitors.
Of course not to imply that OpenAI is the same as FTX.
It was just a rank and file IC.
At this point Gemini is as good as GPT4 (and anecdotally I think it's better at many coding assistance tasks, based on my experiments on the LMSys chatbot arena). Sora is getting a lot of press for text-to-video, but Google's Lumiere has already been out for a few weeks and produces pretty good results.
I have no doubt that OpenAI has been cooking things up. GPT-4 is an older model which others are only just catching up to now. They have an A+ team, a giant war chest, and a lot of momentum. But just because they have momentum does not mean that they have a moat.
I'd be willing to wager that the top-of-the-line foundational models are going to converge and become indistinguishable for almost all tasks. Even the open foundation models (e.g. Mixtral) are getting really good. Foundation models are not moats.
The players that have moats are Microsoft (with deep enterprise software & B2B expertise, allowing them to sell AI-powered software & ward off competition from upstarts) and Google (where their decade of investment into custom silicon allows them to train & run inference for cheaper than anyone else, by far).
10m context with that retrieval rate is such a monstrous leap. And to top it off, we got LargeWorldModel in the same week, capable of 1M token context with insane retrieval rate in the open source space. So not only is the open source world currently technically ahead of ChatGPT, so is Google. Which is why they had to announce SORA, because google's model is so far ahead of the competition. That's also why it will probably be ages before we get access to SORA. Now don't get me wrong, the average person can't afford 32 TPU's to run LWM, but we already have quants for it, which is a step towards enabling the average person (that somehow has 24-48gb of VRAM to get a taste of that power).
What is also striking is the fact that the new models are all multimodal as a standard. We not only leapfrogged in context size, but also in modalities. The model seems to only benefit from having more modalities to work with.
I think the statement Bill Gates made claiming that "LLM's have reached a plateau" itself indicates they don't believe they can make more money from training better/larger models. Which indicates that they already did as well as they could with their existing people, and are now "years" behind google. I never thought google could catch up, especially after their infamous "We have no moat" situation. But it seems they actually doubled down and did something about it.
To a lot of people, last Thursday was a very nihilistic day for Local Models, as the goalposts shifted from 128-200k context to 10M tokens with near perfect retrieval. It's literally insanely scary. But luckily we got LWM, and that means we have only been 10xed.
Now the local people will work on figuring out how to bridge the gap, before being leapfrogged again. What is really insane is that, we have had LLAMA2 for over a year now, and nobody else figured out how to get this result from it, despite it being around so long.
I still believe there are modifications to the architecture of MoE that will unlock new powers that we haven't even dreamed of yet.
Sorry, this was supposed to be well thought out, but it turned more into stream of consciousness, and I honestly had no intention of disagreeing with you.
If I remember the paper correctly, it was something about a 4M context in there. So not 10x, but 2.5x.
> What is really insane is that, we have had LLAMA2 for over a year now, and nobody else figured out how to get this result from it, despite it being around so long.
This isn't true. For now, the task of extending context to 10M tokens is brute-forced by money (increased HW requirements for training and inference and increased training time are also a financial domain). And for now, there simply is no leapfrogging solution for open source or commercial models, which will decrease the costs by orders of magnitude.
I literally have a 24/7 consultant with surface level understand of any topic in human history. This consultant also happens to be an amazing artist for $25 a month that gives me the rights to commercialize any art piece they make for me.
This is extremely hard to beat. This will be extremely hard to beat.
As a business they are succeeding.
LLMs are useful tools, but if they don't lead to something you can actually somewhat rely on to generate correct/consistent/factual results (see e.g. the recent Air Canada chatbot lawsuit) then the hype is a bust. TBD.
The bar I'm setting is far from "impossible"; even human children generally won't seamlessly confabulate when you ask them a question they don't know the answer to. Again citing the recent Air Canada case, these models can't even reliably answer simple questions that are definitively and objectively answered in documentation that is presumably made as freely available to them as is technically possible under the limitations of current technology.
No well-informed sources are saying that. If you're saying that mainstream reporting and other non-tech folks are wrong about what the possibilities are then... obviously.
Yes they are, actually! I've talked to people who obsessively read practically every LLM paper that passes through the arxiv, with undeniably deep and broad knowledge of the current state of this tech, who seriously believe it's going to surpass humans within a year or two. That it may already have, in the deep dark top secret labs beneath OpenAI HQ.
However,
> If you're saying that mainstream reporting and other non-tech folks are wrong about what the possibilities are then... obviously.
If it was obvious then why are you still replying to my comments, which have very obviously been specifically addressing the mismatch between hype and reality? If the current approach doesn't scale, the hype will have been a bust! Objectively! That's what I've been talking about this whole time!
"Mainstream reporting and other non-tech folks" is a bit disingenuous, though. The primary drivers of the current unrealistic hype are software vendors and associated clingers-on looking to make a quick buck. They'll say anything, regardless of whether it's true, and as a result of those mostly-falsehoods our public lives will be flooded with awful AI tools that make everything shittier and more difficult. I can't wait!
Like, are you following the things that OpenAI and Deepmind are saying at all? The things that make current LLMs not a threat, they aim to tear down as soon as they can arrange.
OpenAI just released a video network, and one of their core touted benefits was that you could use it as an action controller!
And, um. Do you really think, when the AIs can take a simple prompt and turn it into a ten thousand step plan that requires dynamic skill acquisition, resourcing and persistence, that generating the prompt will be the one single task that stumps them? When we are at that point - and to be clear, every leading AI organisation is sprinting to reach that point earliest - then the difference between doom and safety will be one sentence: "When you are done with that, generate a new prompt." This is not how a world with a long expected lifespan looks.
edit: To be clear, I'm still not accusing OpenAI of making up the doom stuff. Even though when I phrase it like that it sounds like they're directly working on things that obviously end the world, which seems contradictory, I don't think they see it like that. To be honest, I can't explain why any doomer works at OpenAI, except in the way that people sometimes move towards explosions and gunfire. I think it's just a bug in the human brain. We want to have the danger in sight.
So I just don't think this captures an important distinction at the limit. If a system can generate a good action plan, turning it into an agent is just plumbing.
Its worth noting that, at least when working in a truly novel and untested industry, whoever is leading the pack would likely be the first to know when they tech hits a dead end.
By no means am I saying that's actually the case, but there is still a real possibility that LLMs and the underlying architecture don't pan out with regards to the company's goal of developing anything resembling an AGI. If there is a core challenge with the architecture that doesn't scale, OpenAI would just run into the wall when everyone else was still MILES back.
If OpenAI features were to freeze at what we have today I would be surprised if the company stayed around without a major pivot.
Again I'm in no on way saying this is actually the case, only a hypothetical since the tech is still very new and we don't know what we don't know.
This isn't really true. GPT4 is incredibly useful right now and being used by lots of business processes without any improvements being needed.
Every incremental improvement (eg Gemini 1.5 huge context window) opens up even more possibilities.
AGI isn't required at all.
The main problem with freezing is the moat disappears. Others are steadily catching up, and on specific benchmarks, even surpassing. Groq.com is insane wrt perf.
(Although we can make inform estimates on OpenAIs cost structure for serving models well enough to guess they are probably breaking even on it, or pretty close to it. Eg listen to the gradient dissent interview with the Together AI founders and it’s clear they are doing the same, and unless you think OpenAI is remarkably worse at serving technology it implies they are in the breaking even ballpark)
And we still don’t need AGI for it to be a great business.
There is, but in one sense, who cares? The technology is already super useful, even if it ends up not being the end all and be all of AGI. That is, there are really only 2 options:
1. LLMs are an important stepping stone on the way to AGI, in which case OpenAI is in a great position as the company with the best LLM.
2. LLMs turn out to be a "local maximum" in the search for AGI. I think that even if that actually is the case, OpenAI is still in a great position with much of the infrastructure, data, expertise, etc. even if they require totally new model approaches.
Also, I wouldn't worry so much about #2 because if we do ever get AGI, our economy would surely collapse, or else it would look completely unrecognizable to our current economic systems. That is, I'm always struck that when people talk about AGI fears they talk about things like Skynet and misinformation, but nearly our entire society is organized around people selling their labor for a price. I don't know what our economy will look like when/if AGI becomes real, but I do know it will change drastically if it means that the vast majority of people won't be able to price their labor more than $0.
We don't actually know this though. Assuming an AGI hasn't yet been developed, we don't know whether LLMs will actually get us there. We know they seem to have more use than previous ML systems, but until we have an AGI we can't say what will get us there.
Further, are we really assuming that developing AGI is either a shared goal or a given regardless of what people would actually want to happen? It sounds like we agree on the fundamental impacts an AGI would have on our current societal structures, do we as a society not get a say in the change? Have we effectively blessed a handful of people working in the private sector to make that decision for everyone? And if so, when do we grapple with the moral questions, like whether an AGI has rights similar to humans, or if unplugging one is murder, etc?
That is, you responded "We don't actually know this though. Assuming an AGI hasn't yet been developed, we don't know whether LLMs will actually get us there." Exactly, we don't know if this is true, in which case if it's false my second bullet point "LLMs turn out to be a 'local maximum' in the search for AGI." is the true statement.
For example, if GPU-based systems are the limiting factor they wouldn't have the edge. If the problem turns out to be in the human skills and background needed to develop an AGI they similarly wouldn't have the advantage.
Wouldn't there have to be a third scenario, where they walked down a path that doesn't pan out at all and requires a fundamental rethink effectively going back to square one?
Of course, a pretty good position doesn't guarantee winning, but GP didn't claim that.
But yeah, there are probably outcomes in which LLMs are a local maximum and ultimately dead end, and OpenAI will have a hard time holding on because the market turns out more competitive. And somebody might beat them to whatever the next important invention is. We'll see.
I mean we all work in an industry where we literally see inferior tech win out all the time, not just in usage but in money made.
The company could always pivot as they learn the limitations of LLMs and potentially find other options, though I would argue that their current valuation is extremely high and based largely on the promise of tech they really haven't developed yet. If roadblocks are hit they could be in a tight spot having to live up to such high expectations.
That said, as much as I'd like to see the business goal of developing an AGI crash and burn, I wouldn't bet against them. I just hope we somehow solve the seemingly unsolvable alignment problem first.
I'd say that's quite a lead.
Maybe this has changed now with new Gemini model, but still that means one year ahead of everyone.
The same was said about Stripe. Imagine buying Stripe shares at a $96BN valuation when it is now down more than 50% from the peak.
I would indeed sell some if I saw that extreme valuation of any private startup.
Secondary markets always jump around because of lower liquidity. I’d wait until after IPO before making any conclusions there.
If you were an early FB employee trying to diversify your wealth base in 2011[1] it turned out to have been a bad decision in retrospect but made sense then.
It's easy to say in retrospect "greater fool" vs "bad decision" but right now it is reasonable to diversify risk and doesn't imply any bad faith.
[1] https://www.theguardian.com/technology/2011/feb/10/facebook-...
Personally, I thought that Facebook had peaked around IPO and would slowly go irrelevant.
a financial advisor would always tell you to unload all shares at each vest since you are holding more shares in the form of future vestings.they would then tell you to diversify it into ETFs in a mean variance optimized portfolio tailored to your risk appetite using standard deviation of historical returns and correlation of underlying basket of goods. this is a solid approach that produces predictable safe returns.
I would rather let it ride since I have insider information at most companies I work for and have a better understanding of future returns on that stock that the average person on the street.
I say, take average advice, receive average returns.
:)
They were the first to market with their LLM chatbot, but zooming out that forced Google's hand in releasing their own, and there aren't any moats here other than "how much copyrighted data you're willing to train on."
Eventually, OpenAI/MS will be the Bing to Google's, well, Google. It's one thing for early adopters to go and seek out ChatGPT, but Google has so many products already, there's no reason to think that anybody would switch to ChatGPT just for a chatbot if they're already using a Google app for something else.
The same is true for the real-estate market: When a corner of the market (eg. apartments in the city) outperforms the rest of the market, you might stop doing what's right for you and start speculating (being in the process of looking for a place to live, I can see several people who lost significant amounts after having bought in 06/07 and sole 10 years later - hopefully for them, they bought the right place to live and did not care about the losses).
Everyone points to money but they are also working on the most exciting technology in recent history. Crippling both that and likely their long term job prospects vs Sam, whom they never got a straight story for why he was being taken out.
People love to over simplify and reduce things while glossing over all the nuance.
nightmare fuel puppies playing in the snow is what will make the world a better place
Ignore the fear mongering or political/military concerns for a moment. There are very real questions that have to be answered, from logistical questions like how we'll recognize when we develop an intelligence or the moral questions like does an AI have rights similar to humans.
Diamond nanobots ripping us apart to repurpose our atoms isn't a terribly useful conversation. But whether anyone can own an AI, whether turning one off is murder, or whether AIs deserve all the same rights and humans absolutely are. IMO those should have been prerequisites before we even considered questions like how to align an AI or risks of their existence, we shouldn't be going down that road at all if we're setting ourselves up to get caught with our pants down.