Pleias co-founder Alexander Doria argues Anthropic's synthetic environment training for Sonnet 5 lacks a long-term competitive moat

Well the more I dig into synthetic environment, the more likely it seems Anthropic overreached. There is nothing mysterious about the technique, seed data breadth helps but is not a long time moat, much better agents will come.