More shocking Anthropic testimony! 😮 The company reportedly plans to put Claude in charge of its entire research process as soon as NEXT YEAR, and an alignment researcher privately says this comes with a 5% risk that Claude will be PLOTTING AGAINST HUMANITY on day 1 of that handoff. What's worse, this alignment researcher considers that threat "not that big of a deal"!?! So negligible that he's planning to switch teams internally! 😱 We could be facing a 1-in-20 chance of a catastrophic misalignment next year, and Anthropic's own safety expert could hardly be bothered! I think I'm starting to understand why @hilbertspaess resigned! This might sound like hearsay, except that it was relayed *on the record* by none other than ex-OpenAI researcher Daniel Kokotajlo (@DKokotajlo) in front of British Parliament. Here's a snapshot from Daniel's testimony: "I'll share an anecdote from talking to someone at Anthropic a few months ago." "A year from now, two years from now, it's Anthropic's plan to essentially put Claude in charge of the entire research process within Anthropic, including the safety aspects of the research process." "I asked this person: 'If and when that happens, what is the probability that Claudes will be plotting against Anthropic and against humanity at that point?'" "This person said: 'It's more like 5%. Not worth worrying about. I'm in fact going to transfer away from the alignment team because 5% is not that big of a deal.'" "I came away from this conversation being like, 'oh my gosh!'" "This person's job is to prevent this sort of thing from happening, and they think the risk is a 1 in 20 chance of it happening in the very near future in a way that would be catastrophic. And they think a 1 in 20 chance is low enough that they shouldn't really worry about it."