Inside the black box of AI: How AI models actually think | Emmanuel Ameisen (Anthropic)
AI Summary
Emmanuel Ameisen of Anthropic discusses interpretability research, revealing how large language models plan, form general representations, and exhibit structures similar to conscious and unconscious processing. The conversation covers why anthropomorphizing AI can be useful, the concept of J space, and how interpretability connects to AI alignment. References Anthropic's research on tracing thoughts and global workspace, along with scaling laws.
Key Facts
Related topics
Seat-Based SaaS Faces Mass Renewal Churn as Buyers Revolt Toward Consumption Pricing
Read full analysis →Hierarchical Text Generation and Local 800B+ MoEs Threaten Cloud Token Hegemony
Read full analysis →The Enterprise AI Bottleneck Shifts from Code Generation to Strategic Judgement Tasks
Read full analysis →Related
What on earth is going on with AI pricing and packaging?
Elena Verna critiques AI pricing strategies, highlighting vague usage metrics across GrokBot, Muse, Lovable, Instinct, and 11Labs. She argues that AI
How to tell if a startup has a moat
Dan Hockenmaier presents a framework for assessing whether a startup has a durable moat. He argues that only four compounding advantages really matter
Why every company now needs to think and operate like a lab team | Josh Woodward (Google Labs, Gemini, AI Studio)
Josh Woodward, head of Google Labs, discusses how companies must operate like lab teams to innovate in the AI era. The podcast covers idea generation,