Source: Bloomberg (paywall)
The exodus of Google DeepMind talent into startups explicitly positioning themselves against large language models suggests the researcher class sees efficiency or capability limits in the current LLM paradigm—or at least believes venture capital will fund alternatives. This matters: these founders have the credibility and resources to attract top engineering talent and institutional capital that might otherwise fund the next Claude or GPT variant, fragmenting the AI startup landscape into genuine technical bets rather than incremental model scaling.