Benjamin Mann
Co-Founder at Anthropic
Co-founder of Anthropic serving as tech lead for product engineering; one of the architects of GPT-3 at OpenAI before leaving over safety concerns; focused on aligning AI to be helpful, harmless, and honest.
Dimension Profile
Key Themes
Episode Summary
Benjamin Mann, co-founder of Anthropic and one of the architects of GPT-3, shares his deeply informed perspective on AGI timelines, AI safety, and why he left OpenAI to build a safety-first AI company. He discusses why scaling laws are accelerating rather than slowing, his economic Turing test for identifying AGI, the existential risks he estimates at 0-10%, and how these concerns shape both Anthropic's mission and his personal worldview.
Leadership Principles
- → Once we get to superintelligence, it will be too late to align the models — safety must be built in before, not after
- → Safety wasn't the top priority at OpenAI, which is why they left to start Anthropic — sometimes you must build the alternative
- → Superintelligence is about keeping God in a box — the stakes of alignment are existential, between 0 and 10% chance of extremely bad outcomes
Notable Quotes
"I think 50th percentile chance of hitting some kind of superintelligence is now like 2028."
— On his timeline prediction for AGI
"We felt like safety wasn't the top priority there. The case for safety has gotten a lot more concrete, so superintelligence is a lot about how do we keep God in a box and not let the God out."
— On why he left OpenAI to co-found Anthropic
"Once we get to superintelligence, it will be too late to align the models. My best granularity forecast for could we have an X-risk or extremely bad outcome is somewhere between 0 and 10%."
— On the urgency of AI alignment work
Want to know how you compare?
Take the Assessment