Zhen-Fei Yin

Predictive world models, not just static feedback, unlock substantial gains in recursive self-improvement across diverse domains, from code to scientific discovery.

Dual-arm robots can generalize to unseen task compositions by combining skill-conditioned parameters and arm-wise attention, achieving 39% success on physical hardware.

This work extends SocioVerse 1.0 into a human-AI co-evolutionary paradigm built from two loops and one infrastructure, and validates SocioVerse2 across three case families and seven case studies, from reproducing canonical agent-based models to modeling policy processes on real records and nowcasting macro-economic indices beyond the response model's knowledge cutoff.

JEPA-Anything, a domain-agnostic framework based on orthogonal predictive factorization (OPF), achieves the lowest one-step and 100-step molecular errors among compared methods in all four systems and supports a common factorized predictive principle across heterogeneous worlds.

The model family shows gains in held-out scientific-code repair and across selected general-purpose benchmarks in code, reasoning, and knowledge, providing evidence of positive transfer from scientific experience to broader capabilities.

This work introduces and releases ScienceBuddy, an interactive scientific research workspace that brings continually improving scientific agents into researchers'everyday workflows and presents case studies of researcher interaction, harness refinement, and model learning, with the benchmark cases spanning four scientific task families.

This work forms a unified approach to capability formation and process-centered evaluation, enabling discovery behavior to be trained, improved, and measured beyond final-answer performance.

LLM research agents consistently lose to a naive moving-average baseline at forecasting scientific trends—not due to poor reasoning, but because asking models to predict the future systematically biases their search policies away from the most recent literature.

MA-VLA achieves robust multi-arm compositional generalization, outperforming existing models by effectively enabling role-agnostic behavior in unseen collaboration scenarios.

Recuris transforms long-horizon task execution by reducing common failures by up to 80% and achieving state-of-the-art success rates across multiple models.

Subtracting information from the student rather than adding it to the teacher can yield performance improvements that rival those achieved with privileged data.

Skill-switching accuracy in LLMs drops significantly on complex tasks, but a new training approach boosts performance from 34.4% to 68.4% on challenging benchmarks.

Agents can exhibit significant performance gains from retained experience, but the pathways to these improvements are often unclear and model-dependent.