Anthropic's IPO draft shows sales of $16.3bn, Claude designs proteins
Anthropic's draft IPO filing, Claude's work on scorpion toxins, SpaceX's AI business, Anthropic's draft safety bill, locked chats in ChatGPT, Meta's Muse app and New York's chatbot rules.
Anthropic’s draft IPO filing shows sales up nearly 15-fold to $16.3bn That’s for the first half of this year, against about $1.1bn a year earlier, the Wall Street Journal reported on Tuesday after seeing the draft. It also shows a loss of about $81bn, though roughly $72bn of that is accounting charges on Anthropic’s convertible debt rather than money out the door. It spent nearly $9bn on computing in the half. Source
Claude designed 17 proteins that latch onto scorpion toxins An internal Claude model worked on its own for the first 24 hours, came up with 318,484 designs, and two outside labs tested 211 of them. The best two latch tightly onto a Mexican scorpion’s toxin, at about the same spot as an antibody reported to neutralise it in mice. Anthropic put it out on Tuesday and says nobody has shown they work against venom in cells or animals. Source
SpaceX’s AI arm nearly tripled its quarterly sales to $7.56bn That’s the unit behind Grok, now called Super Intelligence, and it made up about half of SpaceX’s $14.6bn in revenue for July to September, which was up 161 per cent on a year earlier. SpaceX reported on Tuesday, beat what analysts expected on revenue, and spent $19.6bn on building things out in the quarter. Source
Anthropic wants judges able to block models that could kill 50 Politico reported the draft law on Friday and Anthropic confirmed it. Registered outside testers would get 72 hours to warn the Commerce Department of an imminent risk of more than 50 deaths or over $1bn in damage, from bioweapons, cyberattacks or a model slipping its makers’ control. Commerce would then ask a Washington judge to stop the release, and agencies couldn’t halt a model on their own. Source
ChatGPT now lets you lock chats behind a four-digit PIN Locked chats, which can also open with Face ID, sit in their own folder with their own memory, so nothing in them leaks into your normal chats. It started rolling out on Tuesday on every plan but ChatGPT for Healthcare. It isn’t encryption, and if you forget the PIN, resetting it deletes everything you’d locked. Source
Meta’s Muse app hit 10 million US downloads in 40 days Only 11 other apps have ever done that, Sensor Tower says, and the last one was Threads in 2023. ChatGPT got 3.7 million US downloads in its first 40 days, though Muse is only in the US and Canada, and Meta is putting 39 per cent of its ad spend behind it this month, up from 23 per cent. The numbers came out on Tuesday. Source
New York City wants chatbots to let you reach a human Under rules the city’s consumer protection department proposed on Tuesday, businesses would have to tell you when you’re talking to a bot or an automated phone line, and give you an easy way through to a real person, with fines starting at $525. There’s a 30-day comment period first. Source