Public GitHub pushes doubled in a year as researchers flag AI-generated code feeding back into pretraining
aminkarbasi · x · 2026-11-09
DimitrisPapail shared data showing pushes to public GitHub repos per quarter have doubled in a single year since 2020, quipping "the pretraining tokens are flowing."
Amin Karbasi offered a sharper take: this growth is essentially post-training tokens flowing back into pretraining — massive volumes of AI-assistant-generated code being committed to public repos and becoming training data for future models. Papail replied it's "basically true."
Related event: GitHub Public Repo Pushes Double to 408M Per Quarter in a Year(2 posts)→
More from AGI Musings
- Critics slam AI ban bill's medical exemption: anything that can make drugs can make poisons — TinfoilTricorn · 2026-11-09
- AI-generated, AI-verified proof no human can read sparks unfalsifiable review debate — cloneofsimo · 2026-11-09
- Tencent chief AI scientist Shunyu Yao: AI will soon master biology, plans fully autonomous AI labs — nlarusstone · 2026-11-09
- Small Task-Specific Agents Beat General Intelligence in India, Says Definable AI Founder — anandesh-sharma · 2026-11-09
- Bitter lesson debate: Rich Sutton's two views aren't actually in conflict — allenainie · 2026-11-09
- First/last author incentives are the biggest blocker to interdisciplinary team science — RexDouglass · 2026-11-09