Goodfire launches research series on neural geometry: networks speak English but think in shapes
CatAstro_Piyush · x · 2026-09-26
Interpretability startup Goodfire is publishing a series of posts on its "neural geometry" research agenda, arguing that neural networks might speak English but think in shapes — and that understanding their rich internal geometry is key to precisely debugging and controlling them. The team is also looking for an artist specializing in neural geometry art.
More from Research
- microqwen: hand-coded minimal Transformer ports Qwen 3 modernities into Karpathy's microgpt — anshulkundaje · 2026-09-26
- Bake AI lands two NeurIPS 2026 papers: NARRA-Gym and JobBench with 130 tasks across 35 occupations — my_cat_can_code · 2026-09-26
- Causal writability: low-rank edits can restore physically correct motion in video models — ZimingLiu11 · 2026-09-26
- Stanford prof explains the tech behind a synthetic DNA company: model-guided stochastic synthesis — anshulkundaje · 2026-09-26
- Epoch: Astra Leads in IKEA Furniture Assembly Benchmark — Proper_Actuary2907 · 2026-09-26
- Dataset Distillation project turns an artist's body of work into striking synthetic images — CSProfKGD · 2026-09-26