Induction and Inquiry via Probabilistic Reasoning over Language and Code
Source
arxiv.org
Author
Wasu Top Piriyakulkij, Sam Acquaviva, Cassidy Langenfeld, Joshua Tenenbaum, Kevin Ellis
Date
Context
A computational model of learning combines explanations written in ordinary language with executable code, then revises competing hypotheses as evidence arrives. A LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → helps guide that search, while probabilistic updating represents how strongly the evidence supports each hypothesis. The researchers report that this combination reproduces patterns observed in human learning and information gathering across behavioral studies.
The contribution is a proposed explanation of how people can learn flexible concepts from limited, noisy experience. Matching those behavioral patterns makes the model worth investigating as an account of learning; it does not establish that human brains use the same implementation.
Terms in this piece · Glossary
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
articleModular Cognitive Architecture Emerges in Large Language ModelsPengrui Han, Jacob Andreas, Evelina Fedorenko, Andrea Gregor de Varda
articleRefusing the Impossible: A Taxonomy and Benchmark for Code Hallucination in Large Language ModelsVishnu Asutosh Dasu, Ashish Kundu, Gang Tan
articleCode-as-Auditor: Executable Compliance Reasoning via Regulation-to-CodeJisoo Kim, Taeyoon Kwack, Jinwoo Jang, Woo Kyung Kim, Honguk Woo