Follow-up: substring finding proves surprisingly hard for small neural programs
yuntiandeng · x · 2026-09-22
Follow-up to yuntiandeng's thread on small neural programs: he shares the reproducible examples behind his claim — the May attempt at compiling a function for the strawberry problem, and a July experiment with a similar substring-finding task that turned out to be surprisingly hard. His takeaway stands: reliable simple-task performance likely needs test-time compute and tool use.
Related event: Small neural programs struggle with even trivially simple tasks(2 posts)→
More from Research
- Does sparse attention actually speed up video generation? Three conditions, measured — fruesome · 2026-09-23
- Basecamp Research raises $140m to design drugs from genetic data of 1m+ unstudied organisms — _rockt · 2026-09-23
- Lean's type theory proves Con(ZF) from excluded middle alone — result found with AI — MikePFrank · 2026-09-23
- Mila's Blake Richards wins 2026 Schmidt Sciences Polymath Award for brain-and-algorithms research — Mila_Quebec · 2026-09-23
- When AI Writes DNA: Radical Numerics CEO on the Biosecurity Arms Race — Latent Space · 2026-09-23
- Transformer Explainer: interactive visualization that teaches LLM internals in your browser — blaizedsouza · 2026-09-23