arXiv cs.CL· Harish Tayyar Madabushi, Melissa Torgbi, Claire Bonial·· 13 小时前AI 评分31
论文提出 LLM 通过上下文引导的训练数据先验外推来完成任务，既非随机鹦鹉也非 AGI
Neither Stochastic Parroting nor AGI: LLMs Solve Tasks through Context-Directed Extrapolation from Training Data Priors
AI 导读
一篇立场论文提出，LLM 的许多行为应被解释为“上下文引导外推”——模型从训练数据先验中外推，并用上下文引导其调用合适的先验。作者据此给出可操作的评估清单与设计模式，用于区分记忆、基础外推与真正的应用级泛化，并主张以此校准对 LLM“推理”能力的表述。
正文 · 原文
Abstract:This position paper argues that many behaviours framed as either "stochastic parroting" or "advanced reasoning" are better explained as context-directed extrapolation from training-data priors, and that ML and NLP research communities would benefit from evaluations that distinguish remembering and basic extrapolation from genuine application-level generalisation. Our central contribution is an actionable evaluation checklist and design patterns that make these distinctions testable in practice. Our framework posits that LLMs extrapolate from training data priors while using context to guide the model to the appropriate priors; we call this "context-directed extrapolation." Specifically, this context direction is achieved through examples in base models, leading to in-context learning, while instruction tuning allows analogous task specification via prompts rather than explicit examples. We argue that adopting this more realistic framework and controls for improving evaluation would better calibrate claims about LLM "reasoning," refocus research on how LLM performance interacts with training data and generalisability to out-of-domain tasks, and reduce miscalibration in deployment and policy decisions.
| Comments: | 14 pages, 1 figure (including appendix) |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2505.23323 [cs.CL] |
| (or arXiv:2505.23323v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2505.23323 arXiv-issued DOI via DataCite |
Submission history
From: Claire Bonial [view email]
[v1]
Thu, 29 May 2025 10:31:42 UTC (47 KB)
[v2]
Sat, 20 Sep 2025 13:40:43 UTC (107 KB)
[v3]
Wed, 21 Oct 2026 14:38:37 UTC (89 KB)
来源：arXiv cs.CL · arxiv.org