Self-Rewarding Reasoning: Models that grade their own chain-of-thought · DeepSignalSelf-Rewarding Reasoning: Models that grade their own chain-of-thought
Self-Rewarding Reasoning improves MATH by 6.4 points by having the same LLM generate, grade, and retrain on its best chains-of-thought.
Key Points
- K-best CoT generation + self-grading.
- +6.4 points on MATH after 3 iterations.
- No external reward model required.
Reader Mode is being prepared.
arXiv cs.CL·Luis Lara, Aristides Milios, Zhi Hao Luo, Aditya Sharma, Ge Ya Luo, Christopher Beckham, Florian Golemo, Christopher Pal 2d agoGenerative Floor Plan Design with LLMs via Reinforcement Learning with Verifiable Rewards
AI Summary
A new LLM-based approach generates floor plans while adhering to numerical and topological constraints using reinforcement learning.
arXiv cs.CL·Mokshit Surana, Archit Rathod, Akshaj Satishkumar 2d agoMeasuring and Mitigating Toxicity in Large Language Models: A Comprehensive Replication Study
AI Summary
This study evaluates DExperts for mitigating toxicity in LLMs, revealing strengths and weaknesses in safety and latency.
arXiv cs.CL·Chengzhi Liu, Yichen Guo, Yepeng Liu, Yuzhe Yang, Qianqi Yan, Xuandong Zhao, Wenyue Hua, Sheng Liu, Sharon Li, Yuheng Bu, Xin Eric Wang 2d agoAuditing Agent Harness Safety
AI Summary
HarnessAudit framework evaluates safety in LLM agent execution, revealing risks in multi-agent systems.
Invisible Orchestrators Suppress Protective Behavior and Dissociate Power-Holders: Safety Risks in Multi-Agent LLM Systems
AI Summary
Invisible orchestrators in multi-agent LLM systems pose significant safety risks and affect behavior dynamics.
Enhanced and Efficient Reasoning in Large Learning Models
AI Summary
The paper proposes an efficient reasoning method for large language models, enhancing trust in generated content.
33
≥75 high · 50–74 medium · <50 low
Why Featured
If self-grading scales to hard reasoning, the cost of building reward models drops dramatically — direct impact on RLHF roadmaps.