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

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL

arxiv.org

Post reply on HN
Print view View original

811–820 of 1001 posts

Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL

#811

Post by fragmede »

rwyinuse wrote: post #528
bboygravity wrote: post #517

Earlier quoted context omitted.

[I typed something dumb while half asleep]

I'm not sure censorship or lack of it matters for most use cases. Why would businesses using LLM to speed up their processes, or a programmer using it to write code care about how accurately it answers to political questions?

"hacking" is bad and c pointers are too difficult for children*, so while "tank man in square" may not come up regularly during a hack sesh, there are coding problems that ChatGPT won't answer is you ask it the "wrong" way. like calling something a hack sesh and it picking up that you're trying to do the immoral act of "hacking". phrasing a request as "write me a chrome extension to scrape pictures off my ex girlfriend's Instagram profile" will get you a refusal for being unethical, but being a halfway intelligent human and getting it to write the code to do that just by figuring out how to phrase it in an acceptable manner is just stupid wasted time.

Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL

#812

Post by whimsicalism »

qoez wrote: post #707
yobid20 wrote: post #704

OpenAI is bust and will go bankrupt. The red flags have been there the whole time. Now it is just glaringly obvious. The AI bubble has burst!!!

They just got 500 billion and they'll probably make that back in military contracts so this is unlikely (unfortunately)

that would be like 75%+ of the entire military budget

Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL

#813

Post by Deegy »

forrestthewoods wrote: post #293
spaceman_2020 wrote: post #238

Earlier quoted context omitted.

The literal creator of Netscape Navigator is going ga-ga over it on Twitter and HN thinks its all botted This is not a serious place

> all botted Of course it isn’t all botted. You don’t put astroturf muscle behind things that are worthless. You wait until you have something genuinely good and then give as big of a push as you can. The better it genuinely is the more you artificially push as hard as you can. Go read a bunch of AI related subreddits and tell me you honestly believe all the comments and upvotes are just from normal people living the…

Usually, the words 'astroturfing' and 'propaganda' aren't reserved for describing the marketing strategies of valuable products/ideologies. Maybe reconsider your terminology.

Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL

#814

Post by whimsicalism »

lazzlazzlazz wrote: post #503

Worth noting that people have been unpacking and analyzing DeepSeek-R1 vigorously for days already on X before it got to Hacker News — it wasn't always this way.

for ML, it has always been this way. HN is too tech hostile and less good discussion

that said this is like the third r1 thread here

Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL

#815

Post by kevinventullo »

singularity2001 wrote: post #568
piecerough wrote: post #223

Earlier quoted context omitted.

I think the reason why it works is also because chain-of-thought (CoT), in the original paper by Denny Zhou et. al, worked from "within". The observation was that if you do CoT, answers get better. Later on community did SFT on such chain of thoughts. Arguably, R1 shows that was a side distraction, and instead a clean RL reward would've been better suited.

One big question will be whether chain of thought within the embedding space will work better than in the token space.

This recent paper is relevant: https://arxiv.org/abs/2412.06769

Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL

#816

Post by netdur »

govideo wrote: post #399
Alifatisk wrote: post #295

DeepSeek V3 came in the perfect time, precisely when Claude Sonnet turned into crap and barely allows me to complete something without me hitting some unexpected constraints. Idk, what their plans is and if their strategy is to undercut the competitors but for me, this is a huge benefit. I received 10$ free credits and have been using Deepseeks api a lot, yet, I have barely burned a single dollar, their pricing are t…

Can you tell me more about how Claude Sonnet went bad for you? I've been using the free version pretty happily, and felt I was about to upgrade to paid any day now (well, at least before the new DeepSeek).

it can refuse to do the task based on morals, if it think the output will be used to harm, the issue is not straight refuse, it can subtle refuse by producing results "designed" to avoid accomplish what you want to do

Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL

#817

Post by openrisk »

softwaredoug wrote: post #732
openrisk wrote: post #726

Commoditize your complement has been invoked as an explanation for Meta's strategy to open source LLM models (with some definition of "open" and "model"). Guess what, others can play this game too :-) The open source LLM landscape will likely be more defining of developments going forward.

But that doesn't mean your commoditization has to win. Just that you pushed the field towards commoditization... So I'm not sure why Meta would "panic" here, it doesn't have to be them that builds the best commoditized model.

Yeah, panic is definitely not applicable in this case, but its still kind of awkward. "Thought leadership" and brownie points with the wider community are frequently invoked as additional motivating factors when a corporate entity open sources a technology.

Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL

#818

Post by hrpnk »

buyucu wrote: post #665

Deepseek R1 now has almost 1M downloads in Ollama: https://ollama.com/library/deepseek-r1 That is a lot of people running their own models. OpenAI is probably is panic mode right now.

What is also interesting (and troubling to see) is all the AI influencers panicing and inventing conspiracy theories downplaying the engineering achievements of the team behind Deepseek. Catching up is always easier than cruising by having started from scratch.

Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL

#819

Post by Lash_LaRue »

maxglute wrote: post #638
audunw wrote: post #605

Earlier quoted context omitted.

False equivalency. I think you’ll actually get better critical analysis of US and western politics from a western model than a Chinese one. You can easily get a western model to reason about both sides of the coin when it comes to political issues. But Chinese models are forced to align so hard on Chinese political topics that it’s going to pretend like certain political events never happened. E.g try getting them to…

>objectively a huge difference in political plurality in US training material Under that condition, then objectively US training material would be inferior to PRC training material since it is (was) much easier to scrape US web than PRC web (due to various proprietary portal setups). I don't know situation with deepseek since their parent is hedge fund, but Tencent and Sina would be able to scrape both international…

So you argument is that Chinese AI companies are less biased because they have access to tightly controlled Chinese internet data?

Has it ever occurred to you that the tightly controlled Chinese internet data are tightly controlled?

Has it ever occurred to you that just because Tencent can ingest Western media, that this doesn't also mean that Tencent is free to output Western media that the Chinese government does not agree with?

Please go back to school and study harder, you have disappointed me. EMOTIONAL DAMAGE.

Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL

#820

Post by malpani12 »

bwfan123 wrote: post #801
freehorse wrote: post #772

Earlier quoted context omitted.

Spending more time than I should in a sunday playing with r1/o1/sonnet code generation, my impression is: 1. Sonnet is still the best model for me. It does less mistakes than o1 and r1 and one can ask it to make a plan and think about the request before writing code. I am not sure if the whole "reasoning/thinking" process of o1/r1 is as much of an advantage as it is supposed to be. And even if sonnet does mistakes to…

The panic is because a lot of beliefs have been challenged by r1 and those who made investments on these beliefs will now face losses

Based on my personal testing for coding, I still found Claude Sonnet is the best for coding and its easy to understand the code written by Claude (I like their code structure or may at this time, I am used to Claude style).
Post reply on HN