Emergent temporal abstractions in autoregressive models enable hierarchical reinforcement learning Paper • 2512.20605 • Published 4 days ago • 36
Uncovering mesa-optimization algorithms in Transformers Paper • 2309.05858 • Published Sep 11, 2023 • 13
How do language models learn facts? Dynamics, curricula and hallucinations Paper • 2503.21676 • Published Mar 27 • 1