AI Science Tracker
EN

Ernst Strüngmann Institute

1 discovery from 2026

Discoveries

Tasks solvable separately became harder in a shared recurrent model

Source check: Source checked. Not independent scientific confirmation.

Finding: Across 100 artificial cognitive tasks, individually trained linear recurrent models solved 52 tasks, while a single flat linear recurrent model solved 14. Introducing task-specific features recovered individually solvable linear cases. The comparisons characterize the tested training setup.

Technology: Almost-linear RNNs; fitted hierarchical task features

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

AI contribution
Almost-linear recurrent nets learned individual or jointly shared task dynamics under varied nonlinear capacity.
Test and result
The authors varied nonlinear capacity and separation of task parameters across a synthetic task battery and jointly trained subsets.
Limitation / still unknown
Training failure does not prove mathematical impossibility or establish human or animal learning effects.
Evidence sources and annotation context

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

Evidence sources: https://arxiv.org/abs/2610.33926v1 · https://arxiv.org/html/2610.33926v1

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: v1 architecture, Figure 2 and §4.2.

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

Open the interactive profile for the timeline, or browse all organizations and model families.