Xin Chen

A single open-source model now performs multi-character scene dubbing, in-filling edits, and multi-turn revisions with natural language direction, all at 271 fps.

Benign-looking visual fine-tuning can silently trigger system-wide alignment failures, causing VLMs to hallucinate facts, execute risky agentic actions, and drop jailbreak defenses across completely unrelated tasks.

GeoAAC is proposed, a geometry-based adaptive action chunking method for flow-based VLA policies that adjusts the action horizon according to the reliability of the current action prediction, with geometric variation across action prefixes remaining positively correlated with predictive uncertainty.