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

801–810 of 1001 posts

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

#801

Post by vidarh »

sgt101 wrote: post #774
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…

Can LLM's compute any computable function? I thought that an LLM can approximate any computable function, if the function is within the distribution that it is are trained on. I think it's jolly interesting to think about different axiomizations in this context. Also we know that LLM's can't do a few things - arithmetic, inference & planning are in there. They look like they can because they retrieve discussions from…

> Can LLM's compute any computable function?

In a single run, obviously not any, because it's context window is very limited. With a loop and access to an "API" (or willing conversation partner agreeing to act as one) to operate a Turing tape mechanism? It becomes a question of ability to coax it into complying. It trivially has the ability to carry out every step, and your main challenge becomes to get it to stick to it over and over.

One step "up", you can trivially get GPT4 to symbolically solve fairly complex runs of instructions of languages it can never have seen before if you specify a grammar and then give it a program, with the only real limitation again being getting it to continue to adhere to the instructions for long enough before it starts wanting to take shortcuts.

In other words: It can compute any computable function about as well as a reasonably easily distractable/bored human.

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

#802

Post by column »

synaesthesisx wrote: post #151

Remember, about a month ago Sam posted a comment along the lines of "AI will be capable of superhuman persuasion well before it is superhuman at general intelligence, which may lead to very strange outcomes". The board was likely spooked by the recent breakthroughs (which were most likely achieved by combining transformers with another approach), and hit the panic button. Anything capable of "superhuman persuasion",…

"especially prior to an election cycle"

It looks like you are referring to the USA elections.

1. humanity != USA

2. USA are in a constant election cycle

3. there are always elections coming around the world, so it's never a good time

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

#803

Post by debugnik »

NhanH wrote: post #631
sudosysgen wrote: post #607

Earlier quoted context omitted.

Q* is the optimal (ie, correct for the decision problem) function computing the total expected reward of taking an action from a given state in reinforcement learning.

That is just Q. The asterisk is new

The "star" in A* means optimal (as in actually proven to be optimal, so they could stop publishing A^[whatever] algorithms). I assume either Q* is considered optimal in a way regular Q-learning isn't, or they're mixing Q-learning with some A*-like search algorithm. (Or someone picked an undeserving name.)

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

#804

Post by vidarh »

marshray wrote: post #636
xwolfi wrote: post #597

Earlier quoted context omitted.

You're missing the point: who's using the 'add' instruction ? You. We want 'something' to think about using the 'add' instruction to solve a problem. We want to remove the human from the solution design. It would help us tremendously tbh, just like I don't know, Google map helped me never to have to look for direction ever again ?

When the solution requires arithmetic, one trick is to simply ask GPT to write a Python program to compute the answer. There's your 'add'.

GPT4 now does this by default. You'll see a "analyzing" step before you get the answer, and a link which will show the generated python.

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

#805

Post by vidarh »

davidwritesbugs wrote: post #670
marshray wrote: post #636

Earlier quoted context omitted.

When the solution requires arithmetic, one trick is to simply ask GPT to write a Python program to compute the answer. There's your 'add'.

Interesting, how do you use this idea? If you prompt the LLM "create a python Add function Foo to add a number to another number", "using Foo add 1 and 2", or somesuch, but what's to stop it hallucinating and saying "Sure, let me do that for you, foo 1 and 2 is 347. Please let me know if you need anything else."

With ChatGPT you now just state your problem, and if it looks like math, it will do so. E.g. see this transcript:

https://chat.openai.com/c/dd8de3f7-a50c-4b6d-bd3f-b52ed996d3...

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

#806

Post by cactusplant7374 »

badwolf wrote: post #238

This has been refuted. https://www.theverge.com/2023/11/22/23973354/a-recent-openai...

How can it be a refute when we don’t know the sources? One source vs. another source.

If another article comes out and says the verge article is false will you believe it?

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

#807

Post by ChatGTP »

JacobJeppesen wrote: post #772

Seems like they have made progress in combining reinforcement learning and LLMs. Andrej Karpathy mentions it in his new talk (~38 minutes in) [1], and Ilya Sutskever talks about it in a lecture at MIT (~29 minutes in) [2]. It would be a huge breakthrough to find a proper reward function to train LLMs in a reinforcement learning setup, and to train a model to solve math problems in a similar fashion to how AlphaGo use…

The veil of ignorance has been pushed back and the frontier of discovery forward

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

#808

Post by lewhoo »

seydor wrote: post #730

i no longer follow the messianic complex of the people in openAI. They made great tech, indeed. Other people made great tech before them without instant religious level apocalypse proclamations. People in openAI are smart enough to know that post-AGI , their stock options are worthless anyway so they wouldn't stay walled in their secret garden if such a discovery had been made.

Now I am become Death, the destroyer of worlds.

Apocalypse proclamations aren't just a thing to throw around. Other people made great tech before indeed, but I hope we're not comparing AGI to the next Iphone. There were times in history where development gave us pause. The atomic bomb for one thing, but also superbacteria/viruses, human genetic modifications and perhaps a few others.

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

#809

Post by vidarh »

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…

> More broadly, I can imagine a world where LLM training is a bit more 'interactive'

Well, yes, assume that every conversation you have with ChatGPT without turning off history makes it into the training set.

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

#810

Post by justatdotin »

huitzitziltzin wrote: post #358

Put this on my tombstone after the robots kill me or whatever, but I think all “AI safety” concerns are a wild overreaction totally out of proportion to the actual capabilities of these models. I just haven’t seen anything in the past year which makes me remotely fearful about the future of humanity, including both our continued existence and our continued employment.

given we already have mesh-network surveillance, and autonomous lethal weapons, I'm already feeling unsafe.
Post reply on HN