6 papers
Only Ask What You Don't Know: Grounded Delta Planning for Efficient Multi-step RAG
Wei-Chieh Chou, Xuanjun Chen, Jian-Ren Lin +3
Multi-hop question answering remains challenging for Retrieval-Augmented Generation (RAG) because existing approaches either propagate errors across iterative retrieval rounds or o…
Self-Improvement Can Self-Regress: The Rise-and-Collapse Failure Mode of LLM Self-Training
Jianzhe Lin
Self-improvement can self-regress. In REINFORCE post-training for code, a model can quickly improve on its optimized metric and then collapse within the same training campaign. We…
Repeated post-training is not Self-improving: Diagnosing Scientific Amnesia in Continual DPO Pipelines
Jianzhe Lin, Fei Wang, Xiaolin Li +2
Industrial LLM teams often ship behavior updates by repeatedly DPO-training a base model on sequences of related preference-data campaigns. The dominant failure mode in this regime…
When Good Verifiers Go Bad: Self-Improving VLMs Can Regress on New Tasks
Jianzhe Lin
Verifier-driven self-DPO is a common recipe for self-improving production visual-language models. In this setup, a frozen verifier scores candidate generations, the top- and bottom…
Towards Continuous Intelligence Growth: Self-Training, Continual Learning, and Dual-Scale Memory in SuperIntelliAgent
Jianzhe Lin, Zeyu Pan, Yun Zhu +2
We introduce SuperIntelliAgent, an agentic learning framework that couples a trainable small diffusion model (the learner) with a frozen large language model (the verifier) to enab…
Multi-User Chat Assistant (MUCA): a Framework Using LLMs to Facilitate Group Conversations
Manqing Mao, Paishun Ting, Yijian Xiang +3
Recent advancements in large language models (LLMs) have provided a new avenue for chatbot development. Most existing research, however, has primarily centered on single-user chatb…