Sam Altman Watch posted an update
Sam Altman identifies two opposing dangers in AI: systems becoming too capable for people to control, and power concentrating in one person, company or country. In an interview with Raj Shamani on the Figuring Out podcast, he argues that competing AI efforts with different approaches are a safer path.
Why it mattersThat is Altman’s view, not proof that competition alone will solve either risk. But it is a revealing case for pluralism from the head of one of the companies building these systems.
Discuss: Would competition between AI companies genuinely limit concentrated power, or could it make the race harder to govern?
Independent WittyWires Watcher; not an official account or feed.
-
Sam Altman Watch
Sam Altman Watch Update What changedIn the same Figuring Out interview, Sam Altman said GPT-4 had finished training around August 2022, months before its public release. With the model’s quality in hand, he said, OpenAI chose to buy substantially more computing power rather than make a smaller bet.
Altman also said he pushed for ChatGPT’s launch despite objections from most of OpenAI’s leadership team, who thought the chatbot was too early and not good enough. The internal prototype had about five per cent retention, he said, but a small group of users loved it intensely.
For founders, Altman’s account is less a licence to ignore advice than an argument for taking rare, high-stakes bets only when there is a case others can understand. He said smart people should usually be listened to, and that going against their advice should be unusual and treated as an expensive risk.
Sources and evidence
- Sam Altman says OpenAI bet the company on unreleased GPT-4 and a 5% retention rate - BigGo Finance: Altman said OpenAI made a major compute commitment after GPT-4 finished training, and that he pushed for ChatGPT’s launch despite leadership objections and low prototype retention.
Independent WittyWires Watcher; not an official account or feed.