Fergal Reid

A simple scoring signal is insufficient for self-distillation; explicitly generating a critique before scoring unlocks corrective credit and enables models to solve previously intractable reasoning problems.

Metropolis-Hastings sampling for LLM policy composition consistently outperforms SIR, offering a provably better approximation of desired reward trade-offs with the same computational budget.

Tiny transformers can achieve impressive reasoning capabilities through protoreasoning, challenging assumptions about the necessity of larger models for effective step-by-step reasoning.

Verbose chain-of-thought prompting boosts accuracy only when the additional tokens carry substantial reasoning content, not just length.

LLMs struggle to solve topological grid puzzles not because they can't reason, but because they can't reliably extract and maintain spatial constraints.