AI Paper Atlas

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

From Static Benchmarks to Dynamic Protocol: Agent-Centric Text Anomaly Detection for Evaluating LLM Reasoning

Seungdong Yoa, Sanghyu Yoon, Suhee Yoon, Dongmin Kim, Ye Seul Sim, Junhyun Lee, Woohyung Lim

ICLR 2026 · Large language models › Agents, code and math · approach: Long-horizon LLM agents · 2 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
add Teacher-Student adaptive difficulty scaling to base problemsATAD (GPT-4o generator) · GPT-4o evaluatedaccuracy (%): 94.29 -> 72.43Improved
add Teacher-Student adaptive difficulty scalingATAD (Gemini-2.0-Flash generator) · GPT-4o evaluatedaccuracy (%): 83.29 -> 44.71Improved
add Orchestrator validation (ablation)ATAD (GPT-4o generator) · GPT-4o problem approval rateapproval rate (%): 38.14 -> 87.14Improved
add Orchestrator validation (ablation)ATAD (GPT-4o generator) · Coherence (1-5), GPT-4o reviewercoherence: 3.71 -> 4.74Improved
add Orchestrator validation (ablation)ATAD (Gemini-2.0-Flash generator) · Problem qualityvalidity/coherence/fairness: 5.00 -> 5.00 validity; 4.97 -> 5.00 coherenceNo clear effect
add Orchestrator validation (ablation)ATAD (GPT-4o generator) · accuracy of evaluatorsaccuracy (%): 68.29 -> 72.43 (GPT-4o)Improved
increase number of generated samples from 350 to 7000ATAD (Gemini-2.0-Flash generator) · GPT-4o-mini accuracyaccuracyNo clear effect
evaluate with newer reasoning model GPT-o3-mini/o4-mini instead of GPT-4oATAD (GPT-4o generator) · Final problemsaccuracy (%): GPT-4o 72.43 vs o3-mini 72.14, o4-mini 72.43No 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