AI Paper Atlas

Research map / Large language models / Agents, code and math

How2Everything: Mining the Web for How-to Procedures to Evaluate and Improve LLMs

Yapei Chang, Kyle Lo, Mohit Iyyer, Luca Soldaini

ICML 2026 · Large language models › Agents, code and math · approach: Mathematical and structural reasoning benchmarks · 0 citations (OpenAlex)

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
RL with HOW2SCORE reward on Qwen3-4B-InstructQwen3-4B-Instruct · HOW2BENCHHOW2SCORE: 30.29 -> 43.52Improved
RL with HOW2SCORE reward on Qwen3-8B-InstructQwen3-8B-Instruct · HOW2BENCHHOW2SCORE: 38.52 -> 48.62Improved
RL with HOW2SCORE reward on Olmo-3-7B-ThinkOlmo-3-7B-Think · HOW2BENCHHOW2SCORE: 27.30 -> 37.89Improved
HOW2SCORE-based RL effect on out-of-domain benchmarksQwen3-4B/8B-Instruct, Olmo-3-7B-Think · 12 out-of-domain benchmarksmean OOD change: +0.52, +0.06, +1.05No clear effect
add reference-calibrated length reward to RL (ablation)HOW2SCORE RL · HOW2BENCHgeneration length ratio / apparent gain: length ratio up to 1.34x-1.53x without it -> about 1.0 with itImproved
evaluate RL gains with external judges (GPT 5, Gemini 2.5 Pro) instead of distilled judgeRL-trained models · HOW2BENCHHOW2SCORENo clear effect
RL on Qwen3-8B, blinded manual evaluationQwen3-8B · 200 HOW2BENCH examplesno-failure rate (%): 32.5% -> 40.0%Improved
add SFT stage on instruct checkpointsinstruction-tuned checkpoints · HOW2BENCHHOW2SCORENo clear effect
start post-training from later Olmo 3 7B pretraining checkpoints (RL gain) (ablation)Olmo 3 7B SFT then RL · HOW2BENCHHOW2SCORE RL gain: 3.56 at 10% -> 20.33 at 100%Improved
start SFT from later Olmo 3 7B pretraining checkpoints (SFT gain) (ablation)Olmo 3 7B SFT · HOW2BENCHHOW2SCORE SFT gain: 3.39 to 5.36No clear effect
RL on all topics vs science-only vs dining-only (ablation)Qwen3-8B RL · HOW2BENCHoverall HOW2SCORE gain: all +10.10; science-only +9.41; dining-only +5.55Mixed
repeat source documents up to 10 times during midtraining (ablation)Olmo 3 7B · midtraining memorization evalHOW2SCORE: 14.0 -> 16.5 (document perplexity 10.4 -> 1.4)No 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