reasoning models
6 mentions across all digests
Reasoning models are a class of large language models that produce explicit chain-of-thought outputs showing step-by-step problem solving, studied for inference optimization strategies (parallel vs. sequential sampling) and as a safety monitoring surface, with current models showing limited ability to deliberately obscure their reasoning.
Reasoning Structure Matters for Safety Alignment of Reasoning Models
How AI models structure their reasoning chains—not just what they reason about—becomes critical to whether safety alignment techniques actually work.
Quantifying and Understanding Uncertainty in Large Reasoning Models
Uncertainty quantification frameworks for large reasoning models enable safer AI deployment by measuring model confidence and reliability at scale.
When Reasoning Models Hurt Behavioral Simulation: A Solver-Sampler Mismatch in Multi-Agent LLM Negotiation
Reasoning models paradoxically hurt multi-agent LLM negotiation due to solver-sampler mismatch — they optimize for solving rather than behavioral sampling.
Understanding Performance Gap Between Parallel and Sequential Sampling in Large Reasoning Models
Parallel sampling in large reasoning models doesn't always beat sequential inference—the gap varies significantly based on task complexity and accuracy requirements, reshaping inference optimization strategy.
FoE: Forest of Errors Makes the First Solution the Best in Large Reasoning Models
Forest of Errors reveals that initial reasoning attempts in large language models typically outperform subsequent refinement attempts, suggesting current multi-try inference strategies may be suboptimal.