Denis Stetskov on Mythryl: “We trained AI on 518 million GitHub repos. Most of them are student…”

We trained AI on 518 million GitHub repos. Most of them are student projects, abandoned prototypes, and unreviewed Stack Overflow pastes. Then we were surprised that the output is mediocre. The model isn't broken. It learned exactly what we taught it. techtrenches.dev/p/ai-is-a-mi...

techtrenches.devWhy AI-Generated Code Gets Worse Over Time, Not BetterCMU tracked 807 repos after Cursor adoption. Complexity up 41%. Warnings up 30%. Copilot output now trains the next model. The feedback loop is already closing.

8/5/26, 11:01 am1 repost0 likes

Post