Zhiheng Xi

This work introduces ReDraft (Reference-Driven Revision and Fine-Tuning), which obtains both from the model's own failures: using an expert response only as a reference, it has the model revise its own incorrect rollout, keeps the revision only if a verifier accepts it, and fine-tunes on what survives.

Atria Dawn Preview is introduced, a foundation agentic language model designed for scientific research and engineering workflows, with the goal of expanding the frontier of agent productivity in the real world.

Sci-MMR is introduced, a benchmark for multi-step evidence-grounded scientific reasoning built on structured argument graphs linking scientific claims, citation-grounded knowledge, visual evidence, and supporting regions, and it is found that current answer-centric benchmarks substantially overestimate the evidence-grounded reasoning capabilities of multimodal research agents.

Low-probability tokens disproportionately influence model updates, and a simple reweighting strategy can significantly enhance performance without sacrificing generalization.

Self-improving search agents thrive when feedback and policy evolution are intertwined, leading to sustained performance gains and reduced hallucinations.

IACM-RL reduces infinite loops and stale context errors by proactively managing dynamic user intents, setting a new standard for robust tool invocation.

NexForge transforms the landscape of LLM training by synthesizing 43.2K tasks, propelling model performance to unprecedented levels without the need for domain-specific infrastructure.

Even state-of-the-art language models struggle significantly in real-world tasks, exposing critical shortcomings in their deployment readiness.

Even the top-performing language models struggle with archive-grounded reasoning, achieving only 59.4% accuracy on a benchmark designed to test their agentic capabilities across diverse workplace documents.

AgentDoG 1.5 proves you can achieve GPT-5.4-level agent safety with open-source models trained on just 1k samples, slashing deployment overhead by two orders of magnitude.

Multi-turn reinforcement learning gets a boost: weighting trajectories by semantic similarity dramatically improves baseline estimation and agent performance in long-document visual QA.

Forget benchmarks: AI can now learn "scientific taste" and propose research ideas with higher potential impact than humans, thanks to a novel reinforcement learning approach using citation data.

RFT's impressive in-domain performance masks surprisingly weak generalization to new environments, highlighting a critical challenge for deploying LLM agents in the real world.

GPT-5's scientific reasoning skills plummet by nearly 50% when tackling multi-step workflows, revealing a critical gap in current LLM agents' ability to orchestrate complex tool use.