Yifan Wu

Memory agents can boost decision-making performance by over 8% in long-horizon tasks by actively managing relevant information instead of passively retrieving it.

Interaction inference from screenshots remains a critical challenge, with top models lagging far behind in functional execution despite high visual fidelity.

Stronger coding agents can achieve higher success rates while requiring fewer user interventions, reshaping our understanding of effective coding assistance.

IF-Beta allows student models to achieve superior performance with significantly less data and compute, challenging the traditional reliance on full datasets in knowledge distillation.

Selective teacher intervention in multi-turn training can boost agent performance by over 13% by mitigating the impact of early errors.

Chunk-level semantic verification in OmniOPD yields a +28.64% boost in math performance over traditional OPD, challenging the reliance on token-level logit matching.

Latent-based collaboration may obscure attack risks, with latent-only attacks capable of severely degrading performance without any visible adversarial text.

As AI agents scale and interact, the Foundation Protocol offers a coordination layer that prioritizes accountability and governance, ensuring that the future of human-AI collaboration remains open and governable.

LLM multi-agent systems can achieve significantly higher accuracy at a fraction of the cost by learning to selectively delegate tasks instead of relying on rigid orchestration.

Stop optimizing generative engines in isolation: MAGEO learns reusable editing strategies that dramatically improve visibility and citation fidelity across diverse engines.

On-policy RL for machine learning engineering agents is now practical, thanks to a synthetic sandbox that slashes execution time by 13x while boosting performance by up to 67%.