WIRED: Recursive's Flint shows AI-built AI is getting closer to real recursive self-improvement
cong_ml · x · 2026-10-22
WIRED tested Flint, an open-weight coding model from Recursive, and argues recursive self-improvement (RSI)—using AI to build AI—is getting closer. CEO Richard Socher says nearly every aspect of the model pipeline was done and improved by the company's Eureka Machine, which devised new base-model training methods, architectures, and chip-specific code. A handful of researchers produced a world-class model in months, with benchmarks suggesting Flint beats similar-size models on coding. The loop is still limited for now, but Recursive bets the benefits of RSI outweigh the risks and wants to extend it to other sciences.
More from Models
- Recursive launches Flint, a local coding model built in under 4 months, hitting 125+ tokens/sec on MacBook Pro — noahdgoodman · 2026-10-22
- Grok's in-tweet chat takes ~20s to first token, dev says Google AI search is faster — BraceSproul · 2026-10-22
- Blogger blind-tests Mistral Voxtral 3 French TTS against ElevenLabs after suspicious internal benchmarks — emmanuelvivier · 2026-10-22
- Datalab's Indra OCR: a 1M-parameter open model that hits 44 pages/s on CPUs, beating PaddleOCR — generativist · 2026-10-22
- Pretraining researcher Yingbo Zhou unveils Flint's first model — ChengleiSi · 2026-10-22
- Kimi K2.7 post-trained on 2,400 coding tasks lifts DeepSWE from 31% to 48.7% — echen · 2026-10-22