The Human Last Read Paper: Agent-Native Research Artifact
Restructuring research papers from narrative PDFs into machine-executable knowledge packages with four interlocking layers, so AI agents can understand, reproduce, and extend research.
Publications
Our research advancing AI Co-Scientists.
Sci-Reasoning: A Dataset Decoding AI Innovation Patterns
A comprehensive dataset that decodes the patterns underlying AI research innovation, enabling systematic analysis of how breakthrough discoveries emerge.
EXP-Bench: Can AI Conduct AI Research Experiments?
A benchmark evaluating whether AI systems can autonomously design, execute, and analyze AI research experiments with scientific rigor.
New News: System-2 Fine-tuning for Robust Integration of New Knowledge
A novel approach to fine-tuning language models that enables robust integration of new information without catastrophic forgetting.
Curie: Toward Rigorous and Automated Scientific Experimentation with AI Agents
An AI agent framework designed for conducting rigorous, reproducible scientific experiments with automated hypothesis testing and analysis.