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

261–270 of 280 posts

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

#261

Post by joe_the_user »

usaar333 wrote: post #76
Stevvo wrote: post #20

"The o3 system demonstrates the first practical, general implementation of a computer adapting to novel unseen problems" Yet, they said when it was announced: "OpenAI shared they trained the o3 we tested on 75% of the Public Training set. They have not shared more details. We have not yet tested the ARC-untrained model to understand how much of the performance is due to ARC-AGI data." These two statements are complet…

No they aren't. Every arc problem is novel - that's why it resisted deep learning for so long (and still does to a degree). We just don't know how much the model seeing what an arc problem is on the first place boosts its ability to solve them - that limited statement is all the author is making.

The ARC prize was created last year. Arc hasn't resisted AI for very long.

See; https://en.wikipedia.org/wiki/Fran%C3%A7ois_Chollet

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

#262

Post by motoboi »

bobxmax wrote: post #239
motoboi wrote: post #215

Earlier quoted context omitted.

Because hallucinations is something that from a distance looks very unimportant, but when looked closely is a structural problem. Some people here live very close to the LLM field. Structural because while a human being can be the judge of an LLM output, a computer (or another LLM) cannot. No amount of error correction is enough to turn an LLM output into a reliable input to another (possible dumb) computer system. W…

Two questions: 1) Which search engine comes with infallible information? 2) Where are LLMs being sold as something different?

1) Current (traditional) search engines are indexes. They point to sources which can be read, analyzed and summarized by the human into information. LLM do the read, analysis and summarization part for the human.

2) chatbots, perplexity search engine, summarization chrome extensions, RAG tools. Those all built over the idea that hallucination is a quirk, a little cog in the machine, a minor inconvenience to be dutifully noted (for legal reasons) but conveniently underestimated.

Most things in life don’t have a compiler that will error on a inexistent python package.

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

#263

Post by dutchbookmaker »

svdr wrote: post #105
dr_dshiv wrote: post #68

> The R1-Zero training process is capable of creating its own internal domain specific language (“DSL”) in token space via RL optimization. Um, what’s that now? Really?

Yeah, if I understand correctly AI will create it's own internal reasoning language through RL. In R1-Zero it was already a strange mix of languages. They corrected that for R1 to make the thinking useful for humans.

Not trying to be ironic but it would be interesting to see what this below would look like in the strange mix form:

"If the model's actions involve generating tokens (like in language models), then optimizing these token outputs to maximize reward could lead the model to develop a consistent, efficient way of using tokens that's specific to the problem domain. This might look like a DSL because the tokens are used in a structured, perhaps abbreviated or symbolic way that's efficient for the task, not necessarily human-readable but effective for the model's internal processing."

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

#264

Post by nycdatasci »

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.

Really? What about drug development? Protein folding?

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

#265

Post by enjo »

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?

I would not really classify them as "bad" actors, but there are definitely real research lines into this. This freakonomics podcast (https://freakonomics.com/podcast/how-to-poison-an-a-i-machin...) is a pretty good interview with Ben Zhao at the University of Chicago. He runs a lab that is attempting to figure out how to trip up model training when copyrighted material is being used.

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

#266

Post by vidarh »

boroboro4 wrote: post #221
vidarh wrote: post #219

Earlier quoted context omitted.

They deliver pre-built full racks. The "reasons" are most likely because it's not cost-effective as what is effective at this point a tech demo, that first becomes cheap to run if you're actually going to use a decent portion of the capacity for a single model.

How many servers in one rack? Let's say 42. How many chips in one server? Let's say 8. It's 336 cards per rack - enough for fp8 70B model weights (and, maybe, kv cache if your requests aren't too long, but probably not really). You need 10 (!) racks to serve one (!) DeepSeek model weights. There is also massive amount complexity arises from operating so many nodes. During short time when Groq hardware appeared on the…

Yes, you need a crazy amount of load for them to make sense. But when you're seeing providers build out whole data centres at a cost of billions, there you have their market.

This is a market where several large Nvidia customers are designing their own chips (e.g. Meta, Amazon, Google) because they're at a scale where it makes sense to try.

Whether it's a market that lets Groq be successful remains to be seen.

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

#267

Post by mtrovo »

visarga wrote: post #189

> R1-Zero removes the human bottleneck I disagree. It only removes the bottleneck to collecting math and code reasoning chains, not in general. The general case requires physical testing not just calculations, otherwise scientists would not need experimental labs. Discovery comes from searching the real world, it's where interesting things happen. The best interface between AI and the world are still humans, the code…

The fact that both systems scored well on ARC AGI 1 shows they can handle unseen challenges without heavy human input, unless I'm missing something about why you see humans as the best interface for real world exploration.

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

#268

Post by bobxmax »

motoboi wrote: post #262
bobxmax wrote: post #239

Earlier quoted context omitted.

Two questions: 1) Which search engine comes with infallible information? 2) Where are LLMs being sold as something different?

1) Current (traditional) search engines are indexes. They point to sources which can be read, analyzed and summarized by the human into information. LLM do the read, analysis and summarization part for the human. 2) chatbots, perplexity search engine, summarization chrome extensions, RAG tools. Those all built over the idea that hallucination is a quirk, a little cog in the machine, a minor inconvenience to be dutifu…

> LLM do the read, analysis and summarization part for the human

No they don't. The human is meant to read, analyze and summarize the output same as they would for search results

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

#269

Post by octacat »

merrywhether wrote: post #217
octacat wrote: post #165

Earlier quoted context omitted.

There are ways to analyze that your contributions make sense from the conversation point of view. Reasoning detects that pretty quickly. To attack you would actually use another AI, to generate non totally random stuff. It still could be detected. I would assume to use data they would have to filter it a lot and correlate between many users. You can detect if the user is the real one and trust their other chats "a bi…

You would have to grade every user on every knowledge axis though. Just because someone is an expert in software doesn’t mean you should believe their takes on medicine, no matter how good faith their model interactions appear. I’d argue that coming up with an automated way to determine the objective truthfulness of information would be among the greatest creations of humanity (basically “solving” philosophy), so thi…

of course. Each user input would be compared with other user input and existing data in the model before. Only legit and cross-referenced data could be used. Other data could still be used but marked as "possible controversial data". Good model should know that controversial data exists too and should distinguish it from the proper scientific data on each topic.
Post reply on HN