00:00Ed, can we end on Anthropic in this news that we got on Friday that the government is limiting their
00:05ability to sell the most advanced models to foreigners, therefore Anthropic's like, well, we're not going to offer any of
00:10this.
00:10Ed, this seems really bad for business.
00:12Like, if you are a business using Anthropic and this idea that your model could be turned off at any
00:17time, that's not exactly a reliable thing you'd want to incorporate into your company.
00:23Yeah, so basically June 12th, the Commerce Department put an export control on Anthropic and say, you know, for MIFOs
00:30and Fable, you know, no foreign national use.
00:33And Anthropic's response is basically like, OK, then no one can use it because we can't police this on a
00:40case-by-case basis.
00:41The user base is too big.
00:43Also, they made the observation like many of their employees are foreign nationals, so now their employees can't use it.
00:49They're framing this as being a misunderstanding, you know, the headlines of this morning, because this is a kind of
00:56in-case-you-missed-it story from Friday.
00:58You know, there are foreign governments now in touch with Anthropic about it, the UK being one case study.
01:04But the technical issue is one that is not yet explained.
01:07There's something called a jailbreak, where through a prompt, a certain coding mechanism, you can go outside the guardrails that
01:15are put on a model and get it to do things that those guardrails are deliberately put in place to
01:20avoid.
01:21The issue is that no one's explained the existence of it and the parameters of it.
01:26So it's ongoing, but the relationship between this leading AI company as well as a very prominent IPO candidate and
01:33the U.S. government is back in sort of a self-trending direction.
Comments