AI Science Tracker
EN

Saved shared notes were associated with better puzzle-solving workflows

Dartmouth College; Allen Institute for AI · 22 Oct 2026 · computer science

Discovery

Saved shared notes were associated with better puzzle-solving workflows

Source check: Source checked. Not independent scientific confirmation.

Finding: An audit of 28,568 language-model workflows found that those using a persistent shared document tended to solve more puzzles correctly. Across ten testable puzzles in held-out searches, the median accuracy difference was 13.5 percentage points.

Technology: GPT-5.4 family workflow agents; LightGBM

Family GPT-5: profile · models page

Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.

AI contribution
Language models contribute executable workflow variants and mine feature hypotheses; a fitted LightGBM regressor checks predictive features before held-out searches are opened.
Test and result
The authors mined features of executable workflow variants and evaluated the shared-memory association in held-out search runs. Six of ten comparable test puzzles had intervals strictly above zero.
Limitation / still unknown
The comparison is observational across searched workflows on selected puzzles. Other choices can correlate with shared memory, and the analysis does not establish a causal advantage on other tasks.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://arxiv.org/abs/2610.31709 · https://arxiv.org/html/2610.31709v1

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: linked separately.

Review date: 21 Nov 2026. Review scope: arXiv v1 workflow search corpus, LightGBM validation design, §V and fixed held-out association Table VI.

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open in the full catalog for filters, charts and related discoveries. Browse computer science discoveries.

Related catalog entries

Links use documented relationships where available, plus shared catalog context. A shared topic or model family does not establish a scientific dependency.