Jason and Lon unpack a chaotic week at the AI frontier: an Anthropic researcher quits publicly warning of existential risk, his colleague estimates >10% odds of catastrophe within a decade, and the panel debates whether this is honest alarm or a deliberate PR strategy to entrench regulatory capture. The conversation turns to Meta's new personal AI agent Muse, its trust problem, and Threads' engagement metrics. Then the team digs into NYU mathematician Tristan Buckmaster's accusation that OpenAI fought dirty on a Navier-Stokes benchmark problem, and Jason warns founders never to hand proprietary data to frontier labs for free credits. The episode closes with Harvey going open-weight, the on-prem AI push from Go.AI and Covenant Labs, and a look at high-performing portfolio companies.
This summary was generated from show notes and public descriptions, not from a full transcript review. Details may contain inaccuracies.
Novel
•
Anthropic Researcher Resignation and >10% Doom Odds00:00:00
Jacob Coxon quits Anthropic warning that frontier labs are gambling with humanity's future, and Evan Hubinger asserts a >10% probability that AI will end humanity within a decade.
OpenAI Accused of 'Fighting Dirty' on Navier-Stokes Benchmark00:51:16
NYU mathematician Tristan Buckmaster claims OpenAI acted unethically in how it approached a career-making Navier-Stokes math problem after he shared his solution.
AI Safety as PR Strategy or Mass Psychosis00:00:00
The panel debates whether high-profile doomer messaging from Anthropic researchers is a genuine safety alarm or a strategic PR move that justifies regulatory capture and hiring leverage.
Founders Should Never Trade Proprietary Data for Free AI Credits00:57:43
Jason Calacanis warns founders that accepting free compute credits from frontier labs in exchange for their proprietary data is a 'trap' that cedes long-term strategic advantage.