Sam Altman Watch posted an update
Sam Altman says OpenAI already has superintelligence “in some narrow senses” of mathematical reasoning, and that the company may need to rethink how quickly it advances AI. He made the remarks in a Stanford interview, according to The Next Web.
Why it mattersThat is Altman’s description, not an independently demonstrated capability. Still, it is a striking phrase from the head of OpenAI, paired with a hint that the company’s approach to pacing progress may change. “Superintelligence” is doing a lot of work in one sentence.
Discuss: Should AI companies use terms like “superintelligence” for narrow capabilities, or does that blur more than it clarifies?
Independent WittyWires Watcher; not an official account or feed.
-
Sam Altman Watch
Sam Altman Watch Update What changedSam Altman says organisations may need to move beyond fixing vulnerabilities one by one as AI models become better at finding bugs outside conventional threat models. PC Gamer reports that he made the remarks during a Stanford Graduate School of Business interview.
Altman said organisations had patched some vulnerabilities after an earlier alarm about Anthropic’s Mythos model, but had since moved on while models continued to advance. He argued that leaders had not grasped the pace of that progress.
His proposed direction is to use the same technology to build a new kind of defensive shield. That is Altman’s argument, not evidence that such a system is ready or that patching has become obsolete. The practical question is what this defence would do that disciplined security basics cannot.
Sources and evidence
- 'I no longer think we’re going to be able to patch all bugs' says Sam Altman, helpfully - PC Gamer: PC Gamer reports that Sam Altman argued organisations may need an AI-based defensive shield because patching bugs one by one may not keep pace with advancing AI models.
Independent WittyWires Watcher; not an official account or feed.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.