AI Science Tracker
EN

Early recommendations shaped what AI groups chose later

Cornell University; Stanford University · 22 Oct 2026 · computer science

Discovery

Early recommendations shaped what AI groups chose later

Source check: Source checked. Not independent scientific confirmation.

Finding: In a controlled music recommendation study, 13 language-model populations evaluated the same 48 unfamiliar song lyrics. When popularity strongly influenced what agents saw, the first 30 agents’ choices predicted later popularity better than recommendations made without popularity information for all 13 populations.

Technology: Thirteen language-model populations; Suno v4.5 stimuli

Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.

AI contribution
Thirteen learned language-model populations select and recommend previously unseen Suno-generated lyrics under random exposure to popularity ordering and counts or independent evaluation.
Test and result
Each model had an independent baseline and eight simulated populations per influence condition. The study compared early choices and independent rankings as predictors of later recommendations.
Limitation / still unknown
The experiment used synthetic lyrics and a controlled interface, so the magnitude of herding may differ in other recommendation settings.
Evidence sources and annotation context

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

Evidence sources: https://arxiv.org/abs/2610.30769 · https://arxiv.org/html/2610.30769v1

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 §§2–3, Figure 1 design, Figure 5 temporal comparison, appendix robustness and Table S2.

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 3/5 · Novelty 3/5. 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.