AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Demystifying Scientific Problem-Solving in LLMs by Probing Knowledge and Reasoning

Alan Li, Yixin Liu, Arpan Sarkar, Doug Downey, Arman Cohan

ICML 2026 · Large language models › Retrieval and knowledge · approach: Knowledge graph reasoning with LLMs

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add DeepSeek-R1 knowledge ingredients in context (ablation)Qwen2.5-7B-Instruct · GPQAaccuracy: 35.27 -> 47.19Improved
add DeepSeek-R1 knowledge ingredients in context (ablation)Llama-3.1-8B-Instruct · GPQAaccuracy: 28.13 -> 43.57Improved
add the model's own knowledge ingredients in context (ablation)Qwen2.5-7B-Instruct · GPQAaccuracy: 35.27 -> 34.24No clear effect
add R1 KIs on top of reasoning fine-tuning (Qwen-BOTH) (ablation)Qwen-BOTH · GPQAaccuracy: 40.81 -> 54.46Improved
add reasoning fine-tuning on math plus STEM traces (ablation)Qwen2.5-7B-Instruct · SciReas-Proaccuracy: 13.97 -> 21.11Improved
add reasoning fine-tuning on math-only traces (ablation)Llama-3.1-8B-Instruct · SciReasaccuracy: 31.25 -> 35.49Improved
use KIs from Qwen-Math instead of base Qwen (ablation)Qwen2.5-7B-Instruct · MMLU-Pro*accuracy: 59.03 -> 63.66Improved
use KIs from Llama-Math instead of base Llama (ablation)Llama-3.1-8B-Instruct · GPQAaccuracy: 29.06 -> 29.69No clear effect
raise reasoning effort from low to higho3-mini · SciReas (avg)accuracy: 56.9 -> 63.7Improved
raise reasoning effort from low to highGemini-2.5-Pro · SciReas (avg)accuracy: 66.6 -> 67.0No clear effect
measure on SciReas-Pro instead of SciReasGPT-5 low vs high effort · SciReas-Proaccuracy gap: 3.01 -> 12.22Improved
fine-tune base models on math reasoning traces (no science data) (ablation)Qwen/Llama math variants · knowledge recall of KIsaccuracy: 72.30 -> 73.02No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy