AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Cognitive Fatigue in Autoregressive Transformers: Formalization and Measurement

Riju Marwah, Ritvik Garimella, Vishal Pallagani, Atishay Jain, Michael Charles Stewart, Amit Sheth

ICML 2026 · Large language models › Language, safety and interpretability · approach: Reasoning mechanisms and failure modes · 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
aggregate attention, drift and entropy rather than using entropy alone (ablation)Fatigue Index · HotpotQAAUROC: 0.954 -> 0.976Improved
aggregate signals rather than using drift alone (ablation)Fatigue Index · HotpotQAAUROC: 0.929 -> 0.976Improved
aggregate signals rather than inverted attention alone (ablation)Fatigue Index · HotpotQAAUROC: 0.307 -> 0.976Improved
add hysteresis thresholds (0.50/0.40) to the alert ruleFI alerting · HotpotQAalert flips per generation: 18.934 -> 1.567Improved
add hysteresis thresholds to the alert ruleFI alerting · TriviaQAalert flips per generation: 20.889 -> 1.433Improved
add hysteresis thresholds to the alert ruleFI alerting · SQuADalert flips per generation: 21.085 -> 1.674Improved
compute FI over full generation instead of first 20 tokens (ablation)FI vs repetition · HotpotQASpearman correlation: 0.425 -> 0.848Improved
compute FI over full generation instead of first 20 tokens (ablation)FI vs repetition · SQuADSpearman correlation: 0.375 -> 0.856Improved
switch prompt style from zero-shot to short chain-of-thoughtFI · seed 42 sequencesmean FI: 0.854 -> 0.858No clear effect
vary seed from 42 to 123FI · zero-shot promptsmean FI: 0.854 -> 0.854No clear effect
increase context length from 192 to 1446 filler tokensFI signals · context-length stressprompt attention: len=192 vs len=1446Hurt
place evidence at front instead of middle or end (ablation)attention to evidence · positional sensitivitymean attention: 5--10x higherImproved

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