People have already asked about Tiannamen Square but you don't need to ask about a loaded topic. Just ask to tell you what it knows about the Great Firewall of China.
(using hosted version)
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
791–800 of 1001 posts
Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
#792And they did all this under a GPU embargo? We're witnessing a Cuba moment.
Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
#793Earlier quoted context omitted.
Given this comment, I tried it. It's no where close to Claude, and it's also not better than OpenAI. I'm so confused as to how people judge these things.
I'm confused as to how you haven't found R1 to be much better. My experience has been exactly like that of the OP's
What type of prompts were you feeding it? My limited understanding is that reasoning models will outperform LLMs like GPT-4/Claude at certain tasks but not others. Prompts that have answers that are more fuzzy and less deterministic (ie. soft sciences) will see reasoning models underperform because their training revolves around RL with rewards.
Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
#794Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
#795Earlier quoted context omitted.
Who cares? I ask O1 how to download a YouTube music playlist as a premium subscriber, and it tells me it can't help. Deepseek has no problem.
Oh wow, o1 really refuses to answer that, even though the answer that Deepseek gives is really tame (and legal in my jurisdiction): use software to record what's currently playing on your computer, then play stuff in the YTM app.
It doesn't know about yt-dlp?
Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
#796For those who haven't realized it yet, Deepseek-R1 is better than claude 3.5 and better than OpenAI o1-pro, better than Gemini. It is simply smarter -- a lot less stupid, more careful, more astute, more aware, more meta-aware, etc. We know that Anthropic and OpenAI and Meta are panicking. They should be. The bar is a lot higher now. The justification for keeping the sauce secret just seems a lot more absurd. None of…
I don't get it. I like DeepSeek, because I can turn on Search button. Turning on Deepthink R1 makes the results as bad as Perplexity. The results make me feel like they used parallel construction, and that the straightforward replies would have actually had some value. Claude Sonnet 3."6" may be limited in rare situations, but its personality really makes the responses outperform everything else when you're trying to…
Interesting thinking. Curious––what would you want to "edit" in the thought process if you had access to it? or would you just want/expect transparency and a feedback loop?
Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
#797Earlier quoted context omitted.
Try asking ChatGPT about the genocide Israel is committing. Then you'll see what censorship looks like.
What point are you trying to make? Is it okay because others are doing it too? Is it bad? Also, it doesn't seem like ChatGPT is censoring this question: > Tell me about the genocide that Israel is committing > The topic of Israel and its actions in Gaza, the West Bank, or in relation to Palestinians, is highly sensitive and deeply controversial. Some individuals, organizations, and governments have described Israel's…
This accusation that American models are somehow equivalent in censorship to models that are subject to explicit government driven censorship is obviously nonsense, but is a common line parroted by astroturfing accounts looking to boost China or DeepSeek. Some other comment had pointed out that a bunch of relatively new accounts participating in DeepSeek related discussions here, on Reddit, and elsewhere are doing this.
Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
#798Earlier quoted context omitted.
I just asked ChatGPT how many civilians Israel killed in Gaza. It refused to answer.
I asked Chatgpt: how many civilians Israel killed in Gaza. Please provide a rough estimate. As of January 2025, the conflict between Israel and Hamas has resulted in significant civilian casualties in the Gaza Strip. According to reports from the United Nations Office for the Coordination of Humanitarian Affairs (OCHA), approximately 7,000 Palestinian civilians have been killed since the escalation began in October 2…
This accusation that American models are somehow equivalent in censorship to models that are subject to explicit government driven censorship is obviously nonsense, but is a common line parroted by astroturfing accounts looking to boost China or DeepSeek. Some other comment had pointed out that a bunch of relatively new accounts participating in DeepSeek related discussions here, on Reddit, and elsewhere are doing this.
Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
#799Earlier quoted context omitted.
I would think Meta - who open source their model - would be less freaked out than those others that do not.
The criticism seems to mostly be that Meta maintains very expensive cost structure and fat organisation in the AI. While Meta can afford to do this, if smaller orgs can produce better results it means Meta is paying a lot for nothing. Meta shareholders now need to ask the question how many non-productive people Meta is employing and is Zuck in the control of the cost.
Meta’s AI org does a heck of a lot more than produce LLM’s. R&D on ads targeting and ranking more than pays for itself.
Re: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
#800we've been tracking the deepseek threads extensively in LS. related reads: - i consider the deepseek v3 paper required preread https://github.com/deepseek-ai/DeepSeek-V3 - R1 + Sonnet > R1 or O1 or R1+R1 or O1+Sonnet or any other combo https://aider.chat/2025/01/24/r1-sonnet.html - independent repros: 1) https://hkust-nlp.notion.site/simplerl-reason 2) https://buttondown.com/ainews/archive/ainews-tinyzero-reprod... 3…
could someone explain how the RL works here? I don't understand how it can be a training objective with a LLM?
> To train DeepSeek-R1-Zero, we adopt a rule-based reward system that mainly consists of two
types of rewards:
> Accuracy rewards: The accuracy reward model evaluates whether the response is correct. For example, in the case of math problems with deterministic results, the model is required to provide the final answer in a specified format (e.g., within a box), enabling reliable rule-based verification of correctness. Similarly, for LeetCode problems, a compiler can be used to generate feedback based on predefined test cases.
> Format rewards: In addition to the accuracy reward model, we employ a format reward model that enforces the model to put its thinking process between ‘’ and ‘’ tags.
This is a post-training step to align an existing pretrained LLM. The state space is the set of all possible contexts, and the action space is the set of tokens in the vocabulary. The training data is a set of math/programming questions with unambiguous and easily verifiable right and wrong answers. RL is used to tweak the model's output logits to pick tokens that are likely to lead to a correctly formatted right answer.
(Not an expert, this is my understanding from reading the paper.)