Podcast: Microsoft researcher hits ~45% on ARC-AGI with tiny recursive networks
ziv_ravid · x · 2026-09-15
New episode of The Information Bottleneck features Alexia J., Principal Researcher at Microsoft and author of "Less is More: Recursive Reasoning with Tiny Networks". Topics include:
- Efficiency and tiny models: her recursive reasoning network scores 45% on ARC-AGI with a fraction of the parameters
- Why LLMs are bad at molecules
- Why puzzles suit recursion
- What she'd do with a trillion dollars
Available on the website, YouTube and podcast apps.
More from Models
- Liquid AI's FTPO training cuts small-model "doom loops" from 10.2% to 1.4% — helloiamleonie · 2026-09-15
- Flam's 26B MoE Falcon model returns first token in 30ms, specialized for Indic languages — testingcatalog · 2026-09-15
- OPEN-1B: the world's first fully auditable 1.6B-parameter transformer training run — benfielding · 2026-09-15
- xAI Reportedly Giving Out $50,000 in Grok API Credits at Random — Kyrannio · 2026-09-15
- Same feature, 3 models: $4 Claude vs $0.50 Tencent Hy3 — the cheapest won — tejasghutukade · 2026-09-15
- Models can solve IMO problems, but can they create one? Idea generation lags badly — pranjalssh · 2026-09-15