Mrinank Sharma, who was leading the Safeguards Research Team at Anthropic, has resigned from the company with one of the wackiest letters you’ll ever read, complete with a citation for the bible on CosmoErotic Humanism. The latter being a new global mythology about how all you need is love, man.
Sharma posted his extraordinary resignation letter on X, the first one of these I’ve ever read that features footnotes (four of them) and ends with a poem. Anyway, after a lot of LinkedIn-brained stuff about how wonderful Anthropic is, we get to the meat of the matter:
Sharma’s last project for Anthropic studied how AI assistants can “distort our humanity” but, while it’s tempting to reach for the Skynet button, more incendiary implication follows—namely, that Anthropic may somehow be deviating from its principles to compete.
Back on X, Sharma says he’ll be “moving back to the UK and letting myself become invisible for a period of time.” Who knows, maybe his stocks just vested or something.
Sharma’s departure comes hot on the heels of some other high-profile departures from Anthropic, including R&D engineer Harsh Mehta, AI scientist Behnam Neyshabur, and AI safety researcher Dylan Scandinaro. Unlike Sharma, they all seem to be staying in and around the AI industry rather than cracking open the Wordsworth.
It’s obviously easy to glom onto the bit about “peril” or the suggestion Anthropic is not quite as angelic as it seems but, honestly, this is such vagueposting I’d almost feel like the killer robots were saving us. One X user makes an incredibly good point about it:
“First resignation letter I’ve ever seen that has main character energy (and footnotes),” says FJzeit on X. “It’s a job. You can terminate your contract in a single sentence. You’ll be forgotten in a week.”
“The AI Safety Resignation Letter is now a distinct literary genre,” says Michal Podlewski. “We already have enough material (Leike, Kokotajlo, Sutskever, Schulman, Christiano) for a sizeable volume. At this rate, we’ll have a ‘Best Of’ anthology by the end of the year.”
Podlewski may well be right. Which would be very funny, as long as I don’t have to read it.
AI,Software#Anthropic #AI039s #safety #lead #quits #epic #vaguepost #claiming #039the #world #peril039 #he039s #039invisible039 #study #poetry1770754372
