# Memory Is Not Always Needed: Characterizing Conditional Memory in Scientific Reasoning

arXiv: https://arxiv.org/abs/2608.23982 · HTML: https://mlsanity.com/paper/2608.23982/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 12 of its 40 references that are in the map.
- First posted on arXiv 25 Aug 2026.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 12 of 12)

- [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023)
- [Language Models (Mostly) Know What They Know](https://mlsanity.com/paper/2207.05221/) (2022)
- [Solving math word problems with process- and outcome-based feedback](https://mlsanity.com/paper/2211.14275/) (2022)
- [From AI for Science to Agentic Science: A Survey on Autonomous Scientific Discovery](https://mlsanity.com/paper/2508.14111/) (2025)
- [Larimar: Large Language Models with Episodic Memory Control](https://mlsanity.com/paper/2403.11901/) (2024)
- [Intern-S1: A Scientific Multimodal Foundation Model](https://mlsanity.com/paper/2508.15763/) (2025)
- [Beyond Chemical QA: Evaluating LLM's Chemical Reasoning with Modular Chemical Operations](https://mlsanity.com/paper/2505.21318/) (2025)
- [Personalized Large Language Model Assistant with Evolving Conditional Memory](https://mlsanity.com/paper/2312.17257/) (2023)
- [BioProBench: A Corpus and Benchmark for Biological Protocol Reasoning in Autonomous Science](https://mlsanity.com/paper/2505.07889/) (2025)
- [MeKi: Memory-based Expert Knowledge Injection for Efficient LLM Scaling](https://mlsanity.com/paper/2602.03359/) (2026)
- [Beyond Conditional Computation: Retrieval-Augmented Genomic Foundation Models with Gengram](https://mlsanity.com/paper/2601.22203/) (2026)
- [Augmenting Molecular Language Models with Local n-gram Memory](https://mlsanity.com/paper/2606.12113/) (2026)

## Built on by (most cited first, 1 of 1)

- [C3M: Cross-Session Multimodal Memory Maintenance for Long-Horizon Tasks](https://mlsanity.com/paper/2609.29735/) (2026)

## Abstract (arXiv)

Scientific reasoning requires language models to retrieve specialized knowledge and incorporate it reliably into multi-step computation. Conditional memory provides an explicit lookup pathway that complements dense neural representations, but its usefulness is inherently input- and computation-dependent: retrieved information may repair missing scientific associations, yet it may also introduce distracting shortcuts or interfere with reasoning that the base model can already perform correctly. In this work, we systematically investigate when, where, and to what extent conditional memory should participate in scientific reasoning. We characterize the scientific knowledge boundary and controlled interventions on memory-enabled knowledge-circuit nodes. Based on these analyses, we propose a Knowledge Boundary-Aware Router that uses task-specific input proxies available before generation to determine whether memory is activated, which layer-stage nodes receive memory signals, and how strongly these signals contribute. Experiments on biological and chemical reasoning benchmarks, covering two backbone families and six task types, show that memory effects vary substantially across inputs, tasks, and injection locations. Compared with static and activation-rate-matched random routing, our approach more consistently preserves beneficial memory contributions while suppressing memory-induced regressions, establishing selective memory allocation as an important principle for reliable scientific reasoning.