2024.11

November 24, 2024 Analysis of Deepseek R1

Super long CoT + rethink + Best of N?may be not

November 23, 2024 Interesting, though…

Why LLMs are Vulnerable in Multilingual Blending Attack?

Natural Language Reinforcement Learning

November 22, 2024 Neel commented on it !!!!

Sparse Feature Circuits: Discovering and Editing Interpretable...


November 21, 2024 This a good paper, really… TMLR is great but why not ICLR?

BTW I am really curious about if this will happen in Diffusion Model e.g. CogVideo (3D full attention), considering the increasing “context” length of these attention-based models.

When Precision Meets Position: BFloat16 Breaks Down RoPE in...


November 20, 2024 Feature Alignment

How far are we from (fully) feature-level alignment? | Notion

How far are we from (fully) feature-level alignment?

2024.10