mechanistic-interpretability
(3)
Chris Olah Wants to Read the Mind We Built
For more than a decade, the Anthropic co-founder has pursued one stubborn question: if neural networks can surprise their makers, can we learn to see what they are actually doing?
Andres Carranza
Andres Carranza is the co-founder and CEO of Luzid, a San Francisco startup building David, an agentic AI copilot that orchestrates SAP and Salesforce implementations. A Stanford dropout and former AI researcher, he interned at NASA, Harvard, MIT and was the youngest quantitative trader in Two Sigma's history before leaving school to build a company that aims to turn months-long enterprise software rollouts into something closer to a software update.
Nelson Elhage
Nelson Elhage is a systems engineer turned AI safety researcher who has left fingerprints across the modern software stack. At Anthropic, he co-authored foundational work on mechanistic interpretability and transformer circuits that shaped how the field understands language models. Before that, he was employee ~30 at Stripe and a founding engineer of Sorbet, the Ruby typechecker now used across one of the world's largest payment platforms. His open-source tools - reptyr, livegrep, and ministrace - are staples in the Linux hacker's toolkit. He blogs at 'Made of Bugs' and runs a Buttondown newsletter on computer systems.