Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Evolving OpenAI's Structure

openai.com

Post reply on HN
Print view View original

721–730 of 746 posts

Re: Evolving OpenAI's Structure

#721

Post by xandrius »

kragen wrote: post #682
xandrius wrote: post #381

Earlier quoted context omitted.

How is an LLM more powerful than any nuclear weapon? Seriously curious.

Well, possibly it isn't. Possibly LLMs are limited in ways that humans aren't, and that's why the staggering advances from GPT-2 to GPT-3 and from GPT-3 to GPT-4 have not continued. Certainly GPT-4 doesn't seem to be more powerful than the largest nuclear weapons. But OpenAI isn't limited to creating LLMs. OpenAI's objective is not to create LLMs but to create artificial general intelligence that is better than human…

It seems that the term LLM and AGI are being muddled here. One is a statistic next text generation and the other is what you're trying to say.

LLMs are great at making you think they are the other but aren't.

Re: Evolving OpenAI's Structure

#722

Post by OtherShrezzing »

Meneth wrote: post #644
OtherShrezzing wrote: post #600

Earlier quoted context omitted.

> Superintelligence poses an existential threat to humanity I disagree at least on this one. I don't see any scenario where superintelligence comes into existence, but is for some reason limited to a mediocrity that puts it in contention with humans. That equilibrium is very narrow, and there's no good reason to believe machine-intelligence would settle there. It's a vanishingly low chance event. It considerably chan…

So you assume a superintelligence, so powerful it would see humans as we see ants, would not destroy our habitat for resources it could use for itself?

More fundamental than that, I assume that a superintelligence on that level wouldn't have resource-contention with humans at all.

Re: Evolving OpenAI's Structure

#723

Post by TobTobXX »

voidspark wrote: post #601
TobTobXX wrote: post #594

Earlier quoted context omitted.

Yes! Sounds like a dream. My value isn't determined by some economic system, but rather by myself. There is so much to do when you don't have to work. Of course, this assumes we actually get to UBI first, and it doesn't create widespread poverty. But even if humanity will have to go through widespread poverty, we'd porbably come out with UBI on the other side (minus a few hundred millions starved). There's so much to…

> There's so much to do, explore and learn. Hobbies, hanging out with friends, reading, etc. That's basically it. Probably no international travel. It will be like a simple retirement on a low income, because in a socialist system the resources must be rationed. This will drive a lot of young ambitious people to insanity. Nothing meaningful for them to achieve. No purpose. Drug use, debauchery, depression, violence,…

> Hobbies, hanging out with friends, reading, etc. That's basically it. > It will be like a simple retirement on a low income [...].

Yes, like retirement but without the old age. Right now I'm studying, so I do live on a very low income. But still, there are so many interesting things! For example, I'm trying to design a vacuum pump to 1mbar to be made of mostly 3d printed parts. Do vacuum pumps exist and can I buy them? Absolutely. But is it still fun to do the whole designing process? You bet. And I can't even start explaining all the things I'm learning.

> This will drive a lot of young ambitious people to insanity.

I teach teenagers in the age where they have to choose their profession. The ones going insane will be the unambitious people, those who just stay on TikTok all day and go to work because what else would they do? The ambitious will always have ideas and projects. And they won't mind creating something that already exists, just because they like the process of it.

We already see this with generative AI. Even though you could generate most of the images you'd want already, people still enjoy the process of painting or photographing. Humans are made to be creative and take pleasure from it, even if it is not economically valuable.

Hell, this is Hacker News. Hacking (in its original sense) was about creativity and problem-solving. Not because it will make you money, but because it was interesting and fun.

Re: Evolving OpenAI's Structure

#724

Post by voidspark »

TobTobXX wrote: post #723
voidspark wrote: post #601

Earlier quoted context omitted.

> There's so much to do, explore and learn. Hobbies, hanging out with friends, reading, etc. That's basically it. Probably no international travel. It will be like a simple retirement on a low income, because in a socialist system the resources must be rationed. This will drive a lot of young ambitious people to insanity. Nothing meaningful for them to achieve. No purpose. Drug use, debauchery, depression, violence,…

> Hobbies, hanging out with friends, reading, etc. That's basically it. > It will be like a simple retirement on a low income [...]. Yes, like retirement but without the old age. Right now I'm studying, so I do live on a very low income. But still, there are so many interesting things! For example, I'm trying to design a vacuum pump to 1mbar to be made of mostly 3d printed parts. Do vacuum pumps exist and can I buy t…

Introverted high IQ nerds are a tiny percentage of the world population. We exist in a tiny bubble here in Hacker News.

I am thinking about society as a whole, how it will affect all types of people and cultures on this planet.

Re: Evolving OpenAI's Structure

#725

Post by voidspark »

icetank wrote: post #677
km144 wrote: post #655

Earlier quoted context omitted.

Isn't the problem precisely that uncertainty though? That we have many data points showing that a rotting banana skin will not spontaneously gain sentience, but we have no clear way to predict the future? And we have no way of knowing the true chance of superintelligence arising from the current path of AI research—the fact that it could be 1-in-100 or 1-in-1e12 or whatever is part of the discussion of uncertainty it…

>And we have no way of knowing the true chance of superintelligence arising from the current path of AI research What makes people think that the future advances in AI will continue to be linear instead of falling of and plateau? Don't all breakthrough technologies develop quickly at the start and then fall of in improvements as all the 'easy' improvements have already been made? In my opinion AI and AGI is like the…

We already have flying cars. They’re called airplanes and helicopters. Those are limited by the laws of physics, so we don’t have antigravity flying vehicles.

In the case of AGI we already know it is physically possible.

Re: Evolving OpenAI's Structure

#726

Post by kragen »

xandrius wrote: post #721
kragen wrote: post #682

Earlier quoted context omitted.

Well, possibly it isn't. Possibly LLMs are limited in ways that humans aren't, and that's why the staggering advances from GPT-2 to GPT-3 and from GPT-3 to GPT-4 have not continued. Certainly GPT-4 doesn't seem to be more powerful than the largest nuclear weapons. But OpenAI isn't limited to creating LLMs. OpenAI's objective is not to create LLMs but to create artificial general intelligence that is better than human…

It seems that the term LLM and AGI are being muddled here. One is a statistic next text generation and the other is what you're trying to say. LLMs are great at making you think they are the other but aren't.

Yes, I think it's unfortunate that you decided to muddle the discussion by introducing the discussion of LLMs in your previous comment, and by conflating them with AGI, in the sense of strongly superhuman AI, which is OpenAI's objective. I worked pretty hard to unmuddle them in my comment.

The questions you are bringing up about the possible limits of the LLM approach are interesting open research questions, and while I really doubt your implicit claim to have resolved them, they are ultimately irrelevant to the topic at hand, which, I will remind you, is the astounding novelty of the situation where

> many companies operating in the public eye are basically stating "We are creating a digital god, an instrument more powerful than any nuclear weapon" and raising billions to do it, and nobody bats an eye...

Note that there is nothing about LLMs in this proposition, and the particular company we're implicitly most focused on—OpenAI—has already developed a number of well-known models that aren't LLMs and plans to keep doing so.

Re: Evolving OpenAI's Structure

#727

Post by anticensor »

croemer wrote: post #653
cs702 wrote: post #62

Earlier quoted context omitted.

The charter of a public-benefit corporation gives the company's board and management a bit of legal cover for making decisions that don't serve to maximize, or may even limit, financial returns to shareholders, when those decisions are made for the benefit of the public.

But the reverse isn't true, right? It doesn't prevent the board from maximizing financial returns even when doing so would harm the "public".

Depends on the specific charter.

Re: Evolving OpenAI's Structure

#728

Post by coryfklein »

Retric wrote: post #667
coryfklein wrote: post #664

Earlier quoted context omitted.

Sounds a little too much like, "It's not AGI today ergo it will never become AGI" Does the current AI give productivity benefits to writing code? Probably. Do OpenAI engineers have exclusive access to more capable models that give them a greater productivity boost than others? Also probably. If one exclusive group gets the benefit of developing AI with a 20% productivity boost compared to others, and they develop a 2…

> Does the current AI give productivity benefits to writing code? Probably. > If one exclusive group gets the benefit of developing AI with a 20% productivity boost compared to others, and they develop a 2.0 that grants them a 25% boost, then a 3.0 with a 30% boost, etc... That’s a bit of a stretch, generative AI is least capable of helping with novel code such as needed to make AGI. If anything I’d expect companies…

The "novel AGI code" probably accounts for Then it looks like Company A spends 90% of time on novel research work (while LLMs do all the busy work) and Company B spends 5% of time on novel research work.

Re: Evolving OpenAI's Structure

#729

Post by Retric »

coryfklein wrote: post #728
Retric wrote: post #667

Earlier quoted context omitted.

> Does the current AI give productivity benefits to writing code? Probably. > If one exclusive group gets the benefit of developing AI with a 20% productivity boost compared to others, and they develop a 2.0 that grants them a 25% boost, then a 3.0 with a 30% boost, etc... That’s a bit of a stretch, generative AI is least capable of helping with novel code such as needed to make AGI. If anything I’d expect companies…

The "novel AGI code" probably accounts for Then it looks like Company A spends 90% of time on novel research work (while LLMs do all the busy work) and Company B spends 5% of time on novel research work.

If it were that simple we’d already have AGI.

Just really think about what you just said, sure spend 5% of the time is on the bits nobody on earth has any idea how to accomplish that’s how people will approach this project. Organizationally the grunt work is a trivial rounding error vs the completely unbound we’ve got no idea how to solve this problems bits.

Re: Evolving OpenAI's Structure

#730

Post by ben_w »

latentsea wrote: post #434
ben_w wrote: post #406

Earlier quoted context omitted.

> Edit: because if "AGI" doesn't mean that... then what means that and only that!? "Agentic AI" means that. Well, to some people, anyway. And even then, people are already arguing about what counts as agency. That's the trouble with new tech, we have to invent words for new stuff that was previously fiction. I wonder, did people argue if "horseless carriages" were really carriages? And "aeroplane" how many argued tha…

Agents can operate in narrow domains too though, so to fit the G part of AGI the agent needs to be non-domain specific. It's kind of a simple enough concept... it's really just something that functions on par with how we do. If you've built that, you've built AGI. If you haven't built that, you've built a very capable system, but not AGI.

> Agents can operate in narrow domains too though, so to fit the G part of AGI the agent needs to be non-domain specific.

"Can", but not "must". The difference between an LLM being harnessed to be a customer service agent, or a code review agent, or a garden planning agent, can be as little as the prompt.

And in any case, the point was that the concept of "completely autonomous agentic intelligence capable of operating on long-term planning horizons" is better described by "agentic AI" than by "AGI".

> It's kind of a simple enough concept... it's really just something that functions on par with how we do.

"On par with us" is binary thinking — humans aren't at the same level as each other.

The problem we have with LLMs is the "I"*, not the "G". The problem we have with AlphaGo and AlphaFold is the "G", not the ultimate performance (which is super-human, an interesting situation given AlphaFold is a mix of Transformer and Diffusion models).

For many domains, getting a degree (or passing some equivalent professional exam) is just the first step, and we have a long way to go from there to being trusted to act competently, let alone independently. Someone who started a 3-year degree just before ChatGPT was released, will now be doing their final exams, and quite a lot of LLMs operate like they have just about scraped through degrees in almost everything — making them wildly superhuman with the G.

The G-ness of an LLM only looks bad when compared to all of humanity collectively; they are wildly more general in their capabilities than any single one of us — there are very few humans who can even name as many languages as ChatGPT speaks, let alone speak them.

* they need too many examples, only some of that can be made up for by the speed difference that lets machines read approximately everything

Post reply on HN