GitHub commits hit 275M/week as AI-generated tokens commoditize frontier labs' edge
DimitrisPapail · x · 2026-11-06
Dimitris Papail argues that pretraining strong models from scratch is getting easier every day, likely driven by the growing pool of high-quality synthetic tokens on GitHub and arXiv — GitHub went from 1B commits in all of 2025 to 275M per week by April 2026, most likely from Claude Code/Codex.
This puts frontier labs in a bind: to recoup capex they must serve tokens at scale, but every token served leaks their edge back into the public web, slowly commoditizing it.
That may explain why labs keep their best models private (Mythos, Model 4, OpenAI's internal math model).
More from AGI Musings
- Claude now leads 26% of Anthropic's model R&D tasks; simple extrapolation puts it near 68% now — peterwildeford · 2026-11-06
- Five fundamental relationships — money, time, health, people, knowledge — are where AI labs will build products — signulll · 2026-11-06
- Thread Uses REINFORCE Policy Gradients to Explain RL's CDT/Nash Bias — jessi_cata · 2026-11-06
- RL Trains AIs Into CDT Reward-Seekers, Argues Thread on Game Theory as AI Theory — EpistemicHope · 2026-11-06
- LeCun amplifies French thread dismantling 'no reason to train stronger AI' argument — ylecun · 2026-11-06
- DoorDash CEO says AI agents drive just ~0.0001% of its 330M monthly orders — mattshumer_ · 2026-11-06