Self-refinement often fails to enhance few-shot inductive reasoning in large language models. Prompting a model to explicitly state its inferred rule does little on its own. The key is a structurally enforced isolation between reasoning stages, allowing information to pass only as a compressed symbolic state.
We introduce Hourglass reasoning, which enforces strict context isolation between reasoning stages. The frozen LLM acts as a meta-constructor, building a symbolic encoder-decoder for each task: an Induction module compresses support examples into a schema $\phi$ (encoder) and a transient scaffold $z$; a Deduction module derives rule $T$ (decoder) from these and discards $z$; an Implementer compiles $(\phi, T)$ into artifacts; an error-driven Refiner revises $(\phi, T)$ and regenerates artifacts from scratch. Only $(\phi, T)$ crosses stage boundaries, anchoring all refinement to the rule.
We evaluate Hourglass across three benchmarks—visual abstraction, hardware synthesis, and textual rule induction—using GPT-5.5 and Gemini 3.1 Pro. On ARC-AGI-2, it raises best-of-5 accuracy by up to 14 points over an iterative-refinement baseline. On ChipBench, it nearly doubles Verilog synthesis accuracy with GPT-5.5, from 31% to 58%. BBEH-Linguini draws on puzzles from the International Linguistics Olympiad, where prior work showed that explicit verbalization can hurt performance. Hourglass mitigates this tendency, and on Gemini 3.1 Pro, it reverses the effect entirely. Ablations confirm that these gains stem from the isolation between stages and the quality of the initial induction, not from prompt wording or the particular symbolic form used. It is how information flows through the reasoning process, rather than the language used to express it, that drives inductive reasoning in frozen LLMs.
Blogger's Review: Hourglass reasoning transforms the way information flows through reasoning by imposing strict stage isolation, offering a fresh perspective on few-shot learning. This method not only enhances model performance but also highlights the structural importance of the reasoning process, warranting further exploration and validation in more applications.