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

OpenAI researchers warned board of AI breakthrough ahead of CEO ouster

reuters.com

Post reply on HN
Print view View original

811–820 of 1001 posts

Re: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster

#811

Post by justatdotin »

dmichulke wrote: post #493
jhbadger wrote: post #380

Earlier quoted context omitted.

Exactly. The rational fear is that they will automate many lower middle class jobs and cause unemployment, not that Terminator was a documentary.

By this logic we should just forbid the wheel. Imagine how many untrained people could work in transport and there would always be demand. So why did the wheel not result in mass unemployment? And factories neither? Certainly it should have happened already but somehow it never did...

anyone can make a wheel.

only a handful of (effectively unaccountable) entities have SOTA AIs, and it's very unlikely for others to catch up.

Re: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster

#812

Post by jacquesm »

dwaltrip wrote: post #613
wbhart wrote: post #565

Earlier quoted context omitted.

How on earth could you evaluate the scaling path with too little information. That's my point. You can't possibly know that a technology can solve a given kind of problem if it can only so far solve a completely different kind of problem which is largely unrelated! Saying that performance on grade-school problems is predictive of performance on complex reasoning tasks (including theorem proving) is like saying that a…

For the current generative AI wave, this is how I understand it: 1. The scaling path is decreased val/test loss during training. 2. We have seen multiples times that large decreases in this loss have resulted in very impressive improvements in model capability across a diverse set of tasks (e.g. gpt-1 through gpt-4, and many other examples). 3. By now, there is tons of robust data demonstrating really nice relationsh…

We're at the point now where the harder problem is obtaining the high quality data you need for the initial training in sufficient quantities.

Re: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster

#813

Post by bambax »

wbhart wrote: post #491

I feel very comfortable saying, as a mathematician, that the ability to solve grade school maths problems would not be at all a predictor of ability to solve real mathematical problems at a research level. The reason LLMs fail at solving mathematical problems is because: 1) they are terrible at arithmetic, 2) they are terrible at algebra, but most importantly, 3) they are terrible at complex reasoning (more specifica…

> 4) they (current LLMs) cannot backtrack when they find that what they already wrote turned out not to lead to a solution, and it is too expensive to give them the thousands of restarts they'd require to randomly guess their way through the problem if you did give them that facility

This sounds like a reward function? If correctly implemented couldn't it enable an LLM to self-learn?

Re: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster

#814

Post by nopromisessir »

cduzz wrote: post #585
white_dragon88 wrote: post #517

Earlier quoted context omitted.

Your son is a goddamn genius.

Maybe; he still needs to finish his damned homework and remember to turn it in. And eat some vegetables.

All those things sound very boring to me.

I can offer no concrete solutions.

However, I have a friend who graduated from high school #1 of a big class and 2 years early. His mom explained that if he made at least a 1400(of 1600) on his SAT, she would buy him a new gaming computer. He then proceeded to make exactly a 1400. No more. No less.

I recommend if you haven't tried already, an iteration to this approach using a sliding scale reward system. Perhaps a gaming pc with nvidia 4060ti up to *insert parental budget* in event of a perfect SAT score.

Ofc this only works if he's a gamer. I feel this type of system can be applied in many areas though. In my view, the clever component his mother applied is that the computer he earned was not just a desirable reward... It was VERY desirable.

My parents also tried this system with me. It didn't work as well. The reward was not sizable enough. It just didn't seem worth it. Too low value. Also, I already had a job and bought my own. My parents were unwilling to budget a sufficient reward. It's gotta be something he more or less is unlikely to be able to get via other means.

Now my friend is a physician. He graduated top of his class from med school. I think he's pretty content with life.

The bored ones can be a little more trouble sometimes. Fun breed though. Best of luck.

Re: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster

#815

Post by ChatGTP »

Closi wrote: post #645
nostrademons wrote: post #593

Earlier quoted context omitted.

What I wonder, as a computer scientist: If you want to solve grade school math problems, why not use an 'add' instruction? It's been around since the 50s, runs a billion times faster than an LLM, every assembly-language programmer knows how to use it, every high-level language has a one-token equivalent, and doesn't hallucinate answers (other than integer overflow). We also know how to solve complex reasoning chains…

Why would we teach kids maths then, when they can use a calculator? It's much easier and faster for them. I believe it's because having a foundational understanding of maths and logic is important when solving other problems, and if you are looking to create an AI that can generally solve all problems it should probably have some intuitive understanding of maths too. i.e. if we want an LLM to be able to solve unsolve…

It's also fun to use your brain I guess, I think we've truly forgotten that life should be about fun.

Watching my kids grow up, they just have fun doing things like trying to crawl, walk or drink. It's not about being the best at it, or the most efficient, it's just about the experience.

Now maths is taught in a boring way, but knowing it can help us lead more enjoable lives. When math is taught in an enjoyable way AND people get results out of it. Well that's glorious.

Re: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster

#816

Post by sfjailbird »

I have a strong suspicion that this is a purposeful leak to hype up OpenAI's product. It wouldnt't be out of character for some of the people involved, and it's the kind of thing that rampant commercial (and valuation) focus can bring with it.

Re: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster

#817

Post by falcor84 »

falcor84 wrote: post #284
fullshark wrote: post #98

Earlier quoted context omitted.

If you think AGI is uncertain, then maybe it's just an improvement on a paradigm that is near its end state: amazing autocomplete.

Enlighten me then - what is this limit point of "amazing autocomplete"?

To clarify, I'm really curious about this question. Is there some limit to autocomplete that falls short of continuing a prompt such as: "The following is a literate programming compendium on how to stimulate the human brain in software ..."

Re: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster

#818

Post by namibj »

vidarh wrote: post #768
nostrademons wrote: post #593

Earlier quoted context omitted.

What I wonder, as a computer scientist: If you want to solve grade school math problems, why not use an 'add' instruction? It's been around since the 50s, runs a billion times faster than an LLM, every assembly-language programmer knows how to use it, every high-level language has a one-token equivalent, and doesn't hallucinate answers (other than integer overflow). We also know how to solve complex reasoning chains…

There's no value in an LLM doing arithmetic for the sake of doing arithmetic with the LLM. There's value in testing an LLMs ability to follow the rules for doing arithmetic that it already knows, because the ability to recognise that a problem matches a set of rules it already knows in part or whole and then applying those rules with precision is likely to generalise to overall far better problem solving abilities. B…

Gaslighting LLMs does wonders. In this case, e.g., priming it by convincing it the tool is either inaccessible/overloaded/laggy, or here perhaps, telling it the python tool computed wrong and can thus not be trusted.

Re: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster

#819

Post by palotasb »

_emna wrote: post #781
AgentMatt wrote: post #758

Earlier quoted context omitted.

> To me, it makes sense to define abs(x) = sqrt(x^2) i.e. ignoring the negative solution enforces the positive result. Why does this make sense to you? You have some notion of what an absolute value should be, on an intuitive or conceptual level, and the mathematical definition you give is consistent with that (in the one dimensional case). Now taking this valid definition for the 1-d case and generalizing that to hi…

> Why does this make sense to you? You have some notion of what an absolute value should be, on an intuitive or conceptual level, and the mathematical definition you give is consistent with that (in the one dimensional case). In my mind abs(x) = x*sign(x) which is why the above formulation seems correct. This formulation is useful, for example, in formulating reflections. > Instead, you can go back to the conceptual…

> abs(x) = x*sign(x)

True in 1 dimension, but not in higher dimensions, because, as you say:

> how would you define sign(x) in a higher dimension?

abs(x) is generally defined as distance of x from zero.

The fact that sqrt(x^2) or x*sign(x) happen to give the same result in 1 dimension doesn't necessarily imply that they can be applied in higher dimensions as-is to result in abs(x) with the same meaning. Although sqrt(x^2) is close, but the way to generalize it is sqrt(sum(x[i]^2)).

Re: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster

#820

Post by qsort »

seanhunter wrote: post #797
qsort wrote: post #761

Earlier quoted context omitted.

As someone with a CS background myself, I don't think this is what GP was talking about. Let's forget for a moment that stuff has to run on an actual machine. If you had to represent a quadratic equation, would you rather write: (a) x^2 + 5x + 4 = 0 (b) the square of the variable plus five times the variable plus four equals zero When you are trying to solve problems with a level of sophistication beyond the toy stuf…

I quite agree and so would Wittgenstein, who (as I understand it) argued that precise language is essential to thought and reasoning[1]. I think one of the key things here is often what we think of as reasoning boils down to taking a problem in the real world and building a model of it using some precise language that we can then apply some set of known tools to deal with. Your example of a quadratic is perfect, beca…

> I quite agree and so would Wittgenstein

It depends on whether you ask him before or after he went camping -- but yeah, I was going for an early-Wittgenstein-esque "natural language makes it way too easy to say stuff that doesn't actually mean anything" (although my argument is much more limited).

> I think this is a fundamental problem with the "chat" style of interaction

The continuation of my argument would be that if the problem is effectively expressible in a formal language, then you likely have way better tools than LLMs to solve it. Tools that solve it every time, with perfect accuracy and near-optimal running time, and critically, tools that allow solutions to be composed arbitrarily.

Alpha Go and NNUE for computer chess, which are often cited for some reason as examples of this brave new science, would be completely worthless without "classical" tree search techniques straight out of the Russel-Norvig.

Hence my conclusion, contra what seems to be the popular opinion, is that these tools are potentially useful for some specific tasks, but make for very bad "universal" tools.

Post reply on HN