The Year in AI Papers
The year/Google DeepMind

Paper 2603.09906

Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs

Zorik Gekhman, Roee Aharoni, Eran Ofek, Mor Geva, Roi Reichart, Jonathan Herzig

Published
Mar 2026
Research lab
Google DeepMind
Citations
9
GitHub
Not linked

01 In brief

Summary

The paper investigates why reasoning improves parametric knowledge recall in LLMs for simple, single-hop factual questions.

Using hybrid models (Gemini-2.5-Flash, Gemini-2.5-Pro, Qwen3-32B) on SimpleQA-Verified and EntityQuestions, the authors find that enabling reasoning substantially expands the model's capability boundary, as measured by pass@k, with gains often doubling at higher k.

Question complexity is a poor predictor of reasoning benefit, indicating gains stem from improved recall rather than multi-hop decomposition.

Controlled experiments identify two mechanisms: (1) a computational buffer effect, where generated reasoning tokens enable latent computation independent of semantic content, evidenced by gains from dummy traces, though this effect plateaus; and (2) factual priming, where recalling topically related facts acts as a semantic bridge, as extracting facts from traces and conditioning on them (even with reasoning off) recovers most gains.

However, hallucinated intermediate facts increase the likelihood of final-answer hallucinations, confirmed by a large-scale audit.

Finally, prioritizing reasoning trajectories with hallucination-free factual statements yields relative accuracy improvements of up to 12.2% on SimpleQA-Verified and 5.1% on EntityQuestions.

02 From the paper

Abstract

While reasoning in LLMs plays a natural role in math, code generation, and multi-hop factual questions, its effect on simple, single-hop factual questions remains unclear. Such questions do not require step-by-step logical decomposition, making the utility of reasoning highly counterintuitive. Nevertheless, we find that enabling reasoning substantially expands the capability boundary of the model's parametric knowledge recall, unlocking correct answers that are otherwise effectively unreachable. Why does reasoning aid parametric knowledge recall when there are no complex reasoning steps to be done? To answer this, we design a series of hypothesis-driven controlled experiments, and identify two key driving mechanisms: (1) a computational buffer effect, where the model uses the generated reasoning tokens to perform latent computation independent of their semantic content; and (2) factual priming, where generating topically related facts acts as a semantic bridge that facilitates correct answer retrieval. Importantly, this latter generative self-retrieval mechanism carries inherent risks: we demonstrate that hallucinating intermediate facts during reasoning increases the likelihood of hallucinations in the final answer. Finally, we show that our insights can be harnessed to directly improve model accuracy by prioritizing reasoning trajectories that contain hallucination-free factual statements.