The ‘grown, not built’ view of AI models
A post shares Anthropic researcher Emmanuel Ameisen’s comparison of interpretability research to biology rather than engineering.
TLDR
Anthropic researcher Emmanuel Ameisen describes AI models as “grown,” not built, according to a post sharing a discussion with him. In a quoted excerpt, he says his work feels closer to biology than engineering. The post says the conversation also covers what it calls his discovery that models plan many tokens ahead, the “J-space,” and how interpretability connects to AI alignment and safety.
Combined views
13.5K
6 Sources, first seen ago
