Yang Deng

The fragmented field of world modeling can now be unified under a "levels x laws" taxonomy, revealing critical gaps in autonomous model revision and decision-centric evaluation.

LLMs have consistent, situation-dependent personality "neurons" that can be steered without training, enabling more adaptable and human-like interactions.

LLM agents can achieve superior performance and generalization in complex tasks while slashing token usage by learning from compact, step-level summaries of subtask progress.