When More Thinking Hurts: Overthinking in LLM Test-Time Compute Scaling (2026 · 2 citations)
Test-time scaling research usually assumes that longer reasoning always gives better answers. This paper tests that assumption directly. It measures how much each extra 500 reasoning tokens adds to accuracy. Two open-weight 32B models, DeepSeek-R1-32B and s1-32B, are run under budget forcing at 32 budgets from 500 to…
DynamicFocalPO: Adaptive Focusing Strategy for Preference Optimization (2026)
Preference optimization methods such as DPO align LLMs on pairs of preferred and rejected answers. However, they are easily disturbed by pairs that the model initially ranks the wrong way. FocalPO reduces the influence of such pairs with a modulating factor p^γ, but it keeps γ constant for the entire run. The authors…
The Retrieval Bottleneck: Scaling Laws for Reinforcement Learning in RAG (2026)
Scaling laws let practitioners plan compute for pre-training and for RL on reasoning tasks, but nobody had checked whether they hold for retrieval-augmented generation (RAG) trained with RL.

Co-authors here

5 co-authors with a page, the most papers together first.

Venues

Where their works here were published, the most works first, with the editions of those that have a page.

Papers of our library cite their works 2 times (of the 149,439 papers whose bibliographies we read); 1 citing paper with a page here →.

Updated 2 November 2026