论文提出 Cognitive LbT 框架，解析大模型"从思考中学习"的四种机制
Cognitive Mapping of Learning by Thinking Mechanisms in Large Language Models
搞 self-improvement 或智能体自学习的人可以看这篇。它把 CoT、Self-Instruct、Tree-of-Thoughts 这些方法统一归到四个认知机制下，还解释了为什么模型崩溃和幻觉扩散是结构性的，不是调参能修好的。比单篇方法论文更有全局视角。
一篇 arXiv 论文提出 Cognitive LbT Framework，将大模型的"从思考中学习"（Learning by Thinking）拆解为四种机制：Reflective Reasoning、Synthetic Memory Formation、Internal Simulation 和 Conceptual Self-Organization，并映射到 LLM 的具体计算过程。
分析覆盖 chain-of-thought、self-consistency、self-reflection、Tree-of-Thoughts、Self-Instruct 等代表性方法，认为这些方法可在功能上近似上述机制。例如 Self-Instruct 从 175 个种子任务生成了约 52,000 条指令，在 Super-NaturalInstructions 上取得 33% 的绝对提升。这些方法能在无新增标注数据的情况下改进部分任务表现，但不一定构成模型参数中的持久学习。
论文进一步指出同一机制也带来幻觉扩散、模型崩溃、概念漂移和高推理成本等失败模式，并归因于四个结构性不对称：生成器与监控者同一、学习闭环缺乏外部 grounding、以符号而非感知运动方式模拟、推理时计算成本。框架给出了 LbT 何时成功或失败的可检验命题。
Cognitive Mapping of Learning by Thinking Mechanisms in Large Language Models
Large language models (LLMs) increasingly exhibit capabilities that resemble human Learning by Thinking (LbT), the acquisition of new insight through internal processes such as explanation and reasoning, without new external input. A systematic account of the cognitive mechanisms behind these capabilities, and of their limits, remains underdeveloped. This paper proposes a Cognitive LbT Framework that maps four mechanisms (Reflective Reasoning, Synthetic Memory Formation, Internal Simulation, and Conceptual Self-Organization) onto computational processes in LLMs. Drawing on representative methods, including chain-of-thought prompting, self-consistency, self-reflection, Tree-of-Thoughts search, Self-Instruct, and self-taught reasoning, the analysis argues that selected methods can functionally approximate aspects of these mechanisms and can improve performance on some tasks without additional labeled examples, although this does not always constitute durable learning in model parameters. For example, Self-Instruct generated approximately 52,000 instructions from 175 seed tasks and reported a 33% absolute improvement on Super-NaturalInstructions (Y. Wang et al., 2023). The same mechanisms also produce characteristic failures, including hallucination diffusion, model collapse, concept drift, and high computational cost. These failures are not isolated technical defects but follow from four structural asymmetries between LLMs and human cognition: the identity of generator and monitor, closed learning loops without external grounding, symbolic rather than sensorimotor simulation, and the cost of inference-time computation. The framework yields testable propositions about when LbT in LLMs should succeed or fail and offers a structured basis for designing more reliable autonomous learning systems.
This report is categorized under AITOP's "Papers" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Learning by Thinking, Self-Instruct, Tree-of-Thoughts, chain-of-thought.
Source attribution: curated from arXiv cs.AI (@Dong Han, Jin Zhang, Kunyue Yang).