Max Nadeau

AI:AM LIVE — September 21, 2026 — Coefficient Giving's Max Nadeau on Project Tailwind and $200K-to-$200M Checks for AI-Safety Founders, Why Alignment Research Resists Metrics, and What Third-Party Auditors Are Actually Allowed to Test — Plus Hiring an Autonomous Business Partner Live On Air and Amazon Banning Meta's Muse

A show planned for fifty-five minutes that ran two hours and twelve, and almost none of it was the show that was planned. Nathan Labenz and Prakash Narayanan open twenty minutes late on audio failures that never fully resolve — Nathan finishes the guest interview from his phone — and spend the hour on Jev, the classification-only model that answers with a calibrated choice instead of a chain of thought; on a paper arguing that agents which can talk to each other beat the same number of agents run independently, including a 1,957-byte MNIST model 20 percent smaller than the best human solution; and on Terence Tao's plea that the field slow down, the pile-on that followed it, and Prakash's argument that a thousand-year pipeline for training human mathematicians died in a month. Nathan's read on Noam Brown's Dwarkesh interview is the line he wants shouted from the rooftops: the researchers themselves have been surprised by the pace. The hosts then put the show's AI cohost Cue on a new bidirectional speech model and let it botch a web search live, and sign up for Andon Labs' waitlist-only Pion — an autonomous business agent they name Dawn, tell to go find sponsorships, and explicitly authorize to send outbound email without review. Max Nadeau, who funds technical AI safety at Coefficient Giving, then spends forty-eight minutes on Project Tailwind: $200K to $200M for people who want to found the safety organizations that don't exist yet, with preseed checks available to one person with a good idea and no team. He is candid about the parts that don't flatter the program — his team is the least metrics-driven at a famously metrics-driven funder, he expects the hit rate to fall as the checks get bigger, state law only obliges third-party auditors to check compliance with policies the companies wrote themselves, and he has no idea whether Anthropic IPO money will ever flow through Coefficient Giving. The hosts close on Amazon banning Meta's Muse shopping agent, Nathan arguing that banning early forfeits the option value and the training data, and Prakash predicting the ban is a three-month exclusivity window for an OpenAI shopping agent.