Learning to Reason with LLMs • Hacker News Board

Re: Learning to Reason with LLMs

#781

Post by OutOfHere » Thu, Sep 12, 2024, 9:56 PM UTC

jazzyjackson wrote: post #448

bartman wrote: post #367

This is incredible. In April I used the standard GPT-4 model via ChatGPT to help me reverse engineer the binary bluetooth protocol used by my kitchen fan to integrate it into Home Assistant. It was helpful in a rubber duck way, but could not determine the pattern used to transmit the remaining runtime of the fan in a certain mode. Initial prompt here [0] I pasted the same prompt into o1-preview and o1-mini and both c…

Isn't there a big "Share" button at the top right of the chatgpt interface? Or are you using another front end?

Even for regular users, the Share button is not always available or functional. It works sometimes, and other times it disappears. For example, since today, I have no Share button at all for chats.

Re: Learning to Reason with LLMs

#782

Post by spoonfeeder006 » Thu, Sep 12, 2024, 9:57 PM UTC

So how is the internal chain of thought represented anyhow? What does it look like when someone sees it?

Re: Learning to Reason with LLMs

#783

Post by andrewchambers » Thu, Sep 12, 2024, 9:57 PM UTC

cs391231 wrote: post #408

Student here. Can someone give me one reason why I should continue in software engineering that isn't denial and hopium?

I don't think programming is any less safe than any other office job tbh. Focus on problem solving and using these tools to your advantage and choose a field you enjoy.

Re: Learning to Reason with LLMs

#784

Post by RandomLensman » Thu, Sep 12, 2024, 9:57 PM UTC

HarHarVeryFunny wrote: post #455

RandomLensman wrote: post #218

How could it fail to solve some maths problems if it has a method for reasoning through things?

It's using tree search (tree of thoughts), driven by some RL-derived heuristics controlling what parts of the practically infinite set of potential responses to explore. How good the responses are will depend on how good these heuristics are.

That doesn't sound like a method for reasoning.

Re: Learning to Reason with LLMs

#785

Post by lossolo » Thu, Sep 12, 2024, 9:58 PM UTC

kmeisthax wrote: post #622

>We believe that a hidden chain of thought presents a unique opportunity for monitoring models. Assuming it is faithful and legible, the hidden chain of thought allows us to "read the mind" of the model and understand its thought process. For example, in the future we may wish to monitor the chain of thought for signs of manipulating the user. However, for this to work the model must have freedom to express its thoug…

It's because there is nothing novel here from an architectural point of view. Again, the secret sauce is only in the training data.

O1 seems like a variant of RLRF https://arxiv.org/abs/2403.14238

Soon you will see similar models from competitors.

Re: Learning to Reason with LLMs

#786

Post by delusional » Thu, Sep 12, 2024, 9:58 PM UTC

Great, yet another step towards the inevitable conclusion. Now I'm not just being asked to outsource my thinking to my computer, but instead to a black box operated by a for-profit company for the benefit of Microsoft. Not only will they not tell me the whole reasoning chain, they wont even tell me how they came up with it.

Tell me, users of this tool. What's even are you? If you've outsourced your thinking to a corporation, what happens to your unique perspective? your blend of circumstance and upbringing? Are you really OK being reduced to meaningless computation and worthless weights. Don't you want to be something more?

Re: Learning to Reason with LLMs

#787

Post by RandomLensman » Thu, Sep 12, 2024, 9:59 PM UTC

hatthew wrote: post #576

RandomLensman wrote: post #218

How could it fail to solve some maths problems if it has a method for reasoning through things?

Because some steps in its reasoning were wrong

I would demand more from machine reasoning, just like we demand an extremely low error rate from machine calculations.

Re: Learning to Reason with LLMs

#788

Post by wrath224 » Thu, Sep 12, 2024, 9:59 PM UTC

Trying this on a few hard problems on PicoGYM and holy heck I'm impressed. I had to give it a hint but that's the same info a human would have. Problem was Sequences (crypto) hard.

https://chatgpt.com/share/66e363d8-5a7c-8000-9a24-8f5eef4451...

Heh... GPT-4o also solved this after I tried and gave it about the same examples. Need to further test but it's promising !

Re: Learning to Reason with LLMs

#789

Post by dhdisjsbshsus » Thu, Sep 12, 2024, 10:00 PM UTC

fakedang wrote: post #592

lumost wrote: post #526

Earlier quoted context omitted.

LLMs perform well on small tasks that are well defined. This definition matches almost every task that a student will work on in school leading to an overestimation of LLM capabiity. LLMs cannot decide what to work on, or manage large bodies of work/code easily. They do not understand the risk of making a change and deploying it to production, or play nicely in autonomous settings. There is going to be a massive amou…

Truth is, LLMs are going to make the coding part super easy, and the ceiling for shit coders like me has just gotten a lot lower because I can just ask it to deliver clean code to me. I feel like the software developer version of an investment banking Managing Director asking my analyst to build me a pitch deck an hour before the meeting.

You mentioned in another comment you’ve used AI to write clean code, but here you mention you’re a “shit coder”. How do you know it’s giving you clean code?

Re: Learning to Reason with LLMs

#790

Post by MattDaEskimo » Thu, Sep 12, 2024, 10:01 PM UTC

weweweoo wrote: post #737

MattDaEskimo wrote: post #693

Earlier quoted context omitted.

Most of these posts are from romantics. Software engineering will be a profession of the past, similar to how industrial jobs hardly exist. If you have a strong intuition with software & programming you may want to shift towards applying AI into already existing solutions.

The question is, why wouldn't nearly all other white collar jobs be professions of the past as well? Does the average MBA or whatever possess some unique knowledge that you couldn't generate with an LLM fed with company data? What is the alternative career path? I think software engineers who also understand business may yet have an advantage over pure business people, who don't understand technology. They should be…

They will be of the past.

This is typical of automation. First, there are numerous workers, then they are reduced to supervisors, then they are gone.

The future of business will be managing AI, so I agree with what you're saying. However most software engineers have a very strong low level understanding of programming. Not a business sense of application