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

An analysis of DeepSeek's R1-Zero and R1

arcprize.org

Post reply on HN
Print view View original

61–70 of 280 posts

Re: An analysis of DeepSeek's R1-Zero and R1

#61

Post by baq »

spyckie2 wrote: post #22

> But now with reasoning systems and verifiers, we can create brand new legitimate data to train on. This can either be done offline where the developer pays to create the data or at inference time where the end user pays! > This is a fascinating shift in economics and suggests there could be a runaway power concentrating moment for AI system developers who have the largest number of paying customers. Those customers…

It doesn't need much. 1 good lucky answer in a 1000 or maybe 10k queries gives you the little exponential kick you need to improve. This is how the hockey stick take off looks like and we're already here - OpenAI has it, now deepseek has it, too. You can be sure others also have it; Anthropic at the very least, they just never announced it officially, but go read what their CEO has been speaking and writing about.

Re: An analysis of DeepSeek's R1-Zero and R1

#62

Post by dr_kiszonka »

aprilthird2021 wrote: post #40
mohsen1 wrote: post #31

Earlier quoted context omitted.

every time you respond to an AI model "no, you got that wrong, do it this way" you provide a very valuable piece of data to train on. With reasoning tokens there is just a lot more of that data to train on now

So if I just pay OpenAI $200/mo, and randomly tell the AI, no that's wrong. I can stop the AI takeover?

You would need a lot of pro accounts! I would be surprised if they didn't use any algorithms for detecting well poisoning.

Re: An analysis of DeepSeek's R1-Zero and R1

#63

Post by drdeca »

Onavo wrote: post #5
anothermathbozo wrote: post #3

The claim is that this removes the human bottleneck (aka SFT or supervised fine tuning) on domains with a verifiable reward. Critically, this verifiable reward is extremely hard to pin down in nearly all domains besides mathematics and computer science.

By verifiable do they mean it in the complexity theory P/NP sense of the word?

The other replies have said what was meant, but I don’t think they’ve explicitly addressed whether or not that is the sense used in the idea of NP.

I would say… it is at least somewhat similar.

A problem in NP might be of the form “For this value of X, does there exist a Y such that q(X,Y)?” for some predicate q and value X, and where when the answer is “yes”, the answer of “yes” can be verified by being given a value Y, and evaluating q(X,Y). (Specifically in the case of 3SAT, X would be a 3CNF formula, Y would be an assignment of values to the variables in the formula, and q(X,Y) would be “the formula X when evaluated with variable assignments Y, results in 'true’.”.)

This is sort of like the task of “Given requirements X that can be checked automatically, produce code Y which satisfies those requirements”, except that in this case the question is specifically asking for Y, not just asking whether such a Y exists, but.. well, often in practice when one wants a solution to a problem in NP, one actually wants the witness, not just whether there exists such a Y, right?

So, I would say there is a substantial similarity, but also a difference.

Re: An analysis of DeepSeek's R1-Zero and R1

#64

Post by aithrowawaycomm »

anothermathbozo wrote: post #3

The claim is that this removes the human bottleneck (aka SFT or supervised fine tuning) on domains with a verifiable reward. Critically, this verifiable reward is extremely hard to pin down in nearly all domains besides mathematics and computer science.

It's also extremely hard to nail down in much of mathematics or computer science!

- is such-and-such theorem deep or shallow?

- is this definition/axiom useful? (there's a big difference between doing compass-straightedge proofs vs. wondering about the parallel postulate)

- more generally, discovering theorems is generally not amenable to verifiable rewards, except in domains where simpler deterministic tools exist (in which case LLMs can likely help reduce the amount of brute forcing)

- is this a good mathematical / software model of a given real-world system?

- is the flexibility of dynamic/gradual typing worth the risk of type errors? is static typing more or less confusing for developers?

- what features should be part of a programming language's syntax? should we opt for lean-and-extensible or batteries-included?

- are we prematurely optimizing this function?

- will this program's memory needs play nicely with Rust's memory model? What architectural decisions do we need to make now to avoid headaches 6 months down the line?

Re: An analysis of DeepSeek's R1-Zero and R1

#65

Post by johnfn »

optimalsolver wrote: post #25
dagelf wrote: post #2

Well o3 scored 75% on AGI-1, R1 and o1 only 25%.... watch this space though....

But can o3 write a symphony? Seriously though, I'd like to hear suggestions on how to automatically evaluate an AI model's creativity, no humans in the loop.

Have you tried suno.ai?

Re: An analysis of DeepSeek's R1-Zero and R1

#66

Post by tsunamifury »

BorisMelnik wrote: post #60
nine_k wrote: post #46

Earlier quoted context omitted.

This assumes that you give honest feedback. Efforts to feed deployed AI models various epistemic poisons abound in the wild.

I am not in this space, question: are there "bad actors" that are known to feed AI models with poisonous information?

If the AI already has a larger knowledge domain space than the user then all users are bad actors. They are just too stupid to know it.

Re: An analysis of DeepSeek's R1-Zero and R1

#67

Post by jl6 »

sheepdestroyer wrote: post #21
mohsen1 wrote: post #7

Earlier quoted context omitted.

with 57 million(!!) tokens

From the article : o3 (low) 75.7% 335K $20 o3 (high) 87.5% 57M $3.4K

$3.4K is about what you might pay a magic circle lawyer for an opinion on a matter. Not saying o3 is an efficient use of resources, just saying that it’s not outlandish that a sufficiently good AI could be worth that kind of money.

Re: An analysis of DeepSeek's R1-Zero and R1

#69

Post by jrsdav »

polishdude20 wrote: post #6

I predict that the future of LLM's when it comes to coding and software creation is in "custom individually tailored apps". Imagine telling an AI agent what app you want, the requirements and all that and it just builds everything needed from backend to frontend, asks for your input on how things should work, clarifying questions etc. It tests the software by compiling and running it reading errors and failed tests a…

I have been trying to imagine something similar, but without all the middleware/distribution layer. You need to do a thing? The LLM just does it and presents the user with the desired experience. Kind of upending the notion that we need "apps" in the first place. It's all materialized, just-in-time style.
Post reply on HN