Doom Debates: AI’s Weird New Phase and Sarah Hastings-Woodhouse on p-Doom

AI may be getting smarter, but Sarah Hastings-Woodhouse says it is also getting profoundly weirder—and that mismatch is easy to miss. This condensed recap of Doom Debates trims the full episode into a quick listen, distilling a wide-ranging conversation between host Liron Shapira and Sarah, an AI safety researcher and writer, on AI regulation, alignment, persuasion, and job automation. You’ll hear why she thinks the UK AI Safety Institute matters, what its testing revealed about models and cyberattack capability, and why she sees today’s AI risk as deeply uncertain, with a p-doom estimate in the 40–50% range. The discussion also covers the Hugging Face incident, recursive self-improvement, and why “turning up the fear dial” may be better than either panic or complacency. Listen now to get the key ideas in minutes.