Failing to Explore: Language Models on Interactive Tasks
in Publications
Language models excel at reasoning, but struggle with exploration. This work shows that in interactive settings, they underperform even simple heuristics.
Your LLM Agents are Temporally Blind: The Misalignment Between Tool Use Decisions and Human Time Perception
in Publications
LLM agents fail to account for elapsed time in multi-turn interactions, leading to misaligned tool-use decisions. TicToc reveals this gap and shows temporal alignment requires post-training—not prompting.
Rapid Network Adaptation: Learning to Adapt Neural Networks Using Test-Time Feedback
in Publications
A framework for fast test-time adaptation that enables neural networks to adjust to distribution shifts using feedback signals—without retraining or iterative optimization.