Tiago Pimentel

tpimentel.bsky.social

04 June 2025 15:20

A string may get 17 times less probability if tokenised as two symbols (e.g., ⟨he, llo⟩) than as one (e.g., ⟨hello⟩)—by an LM trained from scratch in each situation! Our new ACL paper proposes an observational method to estimate this causal effect! Longer thread soon!

Sagnik Mukherjee

sagnikmukherjee.bsky.social

21 May 2025 08:27

🚨 Paper Alert: “RL Finetunes Small Subnetworks in Large Language Models” From DeepSeek V3 Base to DeepSeek R1 Zero, a whopping 86% of parameters were NOT updated during RL training 😮😮 And this isn’t a one-off. The pattern holds across RL algorithms and models. 🧵A Deep Dive

Yong Zheng-Xin (Yong)

yongzx.bsky.social

09 May 2025 19:54

📣 New paper! We observe that reasoning language models finetuned only on English data are capable of zero-shot cross-lingual reasoning through a "quote-and-think" pattern. However, this does not mean they reason the same way across all languages or in new domains. [1/N]

Marco

mcognetta.bsky.social

07 May 2025 18:17

Yeah! Join us! discord.gg/pHS8zVUK