Chen Chen

This work proposes an efficient method to add streaming ASR capabilities to an existing duplex S2S model by introducing a lightweight ASR head in parallel to the agent text head, enabling real-time user transcription while preserving full-duplex conversational capabilities including turn-taking and barge-in handling.

Token-level saliency derived from a model's internal reasoning process can drastically improve the efficiency of chain-of-thought compression, outperforming traditional methods.

JUDGESTEALER reveals that efficient extraction of LLM judging capabilities is possible without extensive querying, achieving remarkable accuracy across multiple evaluation protocols.

LLMs can now be surgically updated to not just remember facts, but also consistently *use* them in reasoning, thanks to a new method that maps and adapts the underlying causal circuits.

Real-world speech disfluencies trip up even the most advanced full-duplex voice agents, exposing critical gaps in self-correction and multi-step reasoning abilities.

Mimicking human cognition, FLAIR lets dialogue models "think while listening," boosting performance without adding latency.