Instructions are all you need: Self-supervised Reinforcement Learning for Instruction Following (2026 · 3 citations)
Language models often fail to satisfy instructions that contain several constraints at once. Existing RL methods for this problem depend on stronger models or human annotators for rewards, and multi-constraint tasks give sparse rewards. The authors propose a self-supervised RL framework in which reward signals come…
From Coarse to Fine: Benchmarking and Reward Modeling for Writing-Centric Generation Tasks (2026)
LLMs still have trouble with writing tasks whose prompts contain many explicit constraints. Existing reward-model benchmarks such as RewardBench and JudgeBench score models only broadly. Current training approaches either depend on a costly LLM judge or on reward models that rate general qualities like fluency. The…

Co-authors here

6 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 3 times (of the 149,439 papers whose bibliographies we read); 2 citing papers with a page here →.

Updated 1 November 2026