🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Sigh, it's a bit of a mess. Let me just give you guys the full nuance in one stream of consciousness since I think we'll continue to get partial interpretations that confuse everyone. All the little things I post need to always be put together in one place. First, I have long

🔥The debate’s been wild: How does the reward in RLVR actually improve LLM reasoning?🤔 🚀Introducing our new paper👇 💡TL;DR: Just penalizing incorrect rollouts❌ — no positive reward needed — can boost LLM reasoning, and sometimes better than PPO/GRPO! 🧵[1/n]

What truly drives reasoning in RLVR? Check out our new paper led by Xinyu Zhu for some fascinating insights and analysis!! 🤩

hmmm if you never push up, you maintain more entropy by not doing excessive sharpening. These guys might be onto something🧐

Surprisingly, we find training only with incorrect traces leads to strong performance 🤯 Even more interesting: it improves model diversity and test-time scaling—while correct traces do the opposite. Check out the 🧵👇

Oh man this is a gorgeous idea. Training *against* negative samples but not towards positive ones maintains entropy in the model, therefore increases pass@high k during RL.

Want powerful reasoning in LLMs without the massive RL training costs? 🤯 Our new paper (led by Siru Ouyang) explores transferring reasoning abilities directly from smaller LMs!🚀🚀

Mengzhou Xia Zhepei Wei Wei-Lin Chen Danqi Chen Yu Meng The arxiv version is out! arxiv.org/abs/2506.01347

The Surprising Effectiveness of Negative Reinforcement in LLM Reasoning Paper: arxiv.org/pdf/2506.01347… Code: github.com/TianHongZXY/RL…

Only ding a model for making mistakes! It gives better results in RL and avoids mode collapse. We still understand so little about RL! But we’re learning. Your science dollars at work.

Excited to share our #ICML25 paper (led by Zhepei Wei) on accelerating LLM decoding! ⚡️ AdaDecode predicts tokens early from intermediate layers 🙅‍♂️No drafter model needed 🪶Just lightweight LM heads ✨Output consistency with standard autoregressive decoding Thread👇

🚀🚀Excited to share our new work on Speculative Decoding by Langlin Huang! We tackle a key limitation in draft models which predict worse tokens at later positions, and present PosS that generates high-quality drafts!

Excited to be at #CVPR2025 this week! I’ll be talking about tool-augmented multimodal reasoning in Thursday’s tutorial. Come say hi if you’re around🍻 ⏰ 1:30–5:00 PM CDT, June 12 📍 Room 107 B, CVPR venue

Excited to share that I joined Google DeepMind as a research scientist recently 🥳 Looking forward to future collaborations on exciting projects :)

I wrote a short post on some etiquette for a seemingly mundane task: declining an offer (for a job, internship, grad school, etc). Link in next tweet. 1/2

🚀 Finally live on arXiv after 3 weeks on hold 🤣 Check it out 👉 arxiv.org/abs/2506.15710 #Reasoning #reinforcementlearning #LLMs

Are AI scientists already better than human researchers? We recruited 43 PhD students to spend 3 months executing research ideas proposed by an LLM agent vs human experts. Main finding: LLM ideas result in worse projects than human ideas.

Will be at #ICML2025 next week! We'll present the following works: 🛠️ LarPO: Tue 7/15 (Poster Session 1 East) 🚀 AdaDecode: Wed 7/16 (Poster Session 3 East) 🧮 Negative Reinforcement for Reasoning: Fri 7/18 (AI for Math Workshop) Happy to chat about latest research in LLMs🤩

Thrilled to present three works at #ICML2025!🥳 🚀AdaDecode — Wed 7/16, East Exhibition Hall A-B (#E-2605) 🔢Negative Reinforcement for Reasoning — Fri 7/18, AI for Math Workshop 🤖WebAgent-R1 — Sat 7/19, Workshop on Computer Use Agents Feel free to stop by and chat about #LLMs!