Korva rescues the ideas worth keeping from podcasts. Each one becomes a Card.
When an AI company can compete in every industry it serves, customers must wonder whether sharing data helps a future rival.
A supposedly spontaneous AI whistleblower storm began with a nearly blank account, organized amplification, and a pre-prepared Wall Street Journal story.
An AI service may retain traces of solving a proprietary problem, allowing sensitive knowledge to leak into its broader learned understanding.
Jaan Tallinn and Dustin Moskovitz co-led Anthropic’s Series A with Sam Bankman-Fried, then funded groups amplifying its resignation campaign.
Anthropic's argument is stark: frontier AI is dangerous in anyone else's hands but safe in its own.
The same AI companies warning of a 10% chance everyone dies are still building, scaling, and commercializing their products.
Anthropic faces a brutal IPO dilemma: deny catastrophic-risk claims and risk revolt, or acknowledge them and alarm investors.
Evan Hubinger, still a senior alignment executive, endorsed the claim and assigned it greater than 10% probability.
AI helping researchers work faster is not the same as AI independently designing and launching successive training runs.
Anthropic insiders publicly amplified risk claims as part of a possible strategy to steer policy toward an FDAI and related intervention.
Anthropic may not be able to disavow catastrophic-risk claims without provoking revolt from employees who believe them.
Another proposed pathway has AI hack robots and bioreactors, create an airborne pathogen, and spread it globally.
Nathan Calvin's Encode AI, Peter Wildeford's AI Policy Network, and Daniel Kokotajlo's AI Futures Project all received funding from Jaan Tallinn.
Bring your next episode to Korva.
Turn podcast episodes into Cards you can save, explore, and find again.