Zhi-Qin John Xu

Limit Analysis for Symbolic Multi-step Reasoning Tasks with Information Propagation Rules Based on Transformers

WebSailor-V2: Bridging the Chasm to Proprietary Agents via Synthetic Data and Scalable Reinforcement Learning

Adaptive Preconditioners Trigger Loss Spikes in Adam

Uncritical tokens are 'critical' in pretraining: the implicit regularization effect of Next token prediction

Probability Signature: Bridging Data Semantics and Embedding Structure in Language Models

Identity Bridge: Enabling Implicit Reasoning via Shared Latent Memory

WebSailor: Navigating Super-human Reasoning for Web Agent

An Analysis for Reasoning Bias of Language Models with Small Initialization

Achilles' Heel of Mamba: Essential difficulties of the Mamba architecture demonstrated by synthetic data

The Buffer Mechanism for Multi-Step Information Reasoning in Language Models