Transkript (en)

Anthropic launches Claude Duet. So they opened a research preview of Claude Duet, which is a new app where Claude acts as a proactive assistant. It flags what needs your attention, finishes tasks and keeps working while you're away. It checks with the user before acting on their behalf. All right. So they've got a little bit of a proactive agent situation. Let's see if there's music. Yeah. That, The tools are getting pretty... Or the tools that we're getting to see. Again, I really don't think that these are... I think we've had these for a long time. But they're getting pretty damn insane. The context windows in which it can deal with too. And the ease of access of the agents are pretty insane. So they still do make mistakes. I think people will always go to the fact of... It can't do it perfectly. it's like you're totally right but very often it's human error in terms of the lack of context given in the prompt itself or the communication there and then in turn i not even saying that it gets to a point where it could do everything you want it to because it impossible it has like no like like be in your brain which hopefully don get there but i say even today it at a point where you can give it like a day worth of tasks it go work for like an hour and then you can come back and just see what it did and approve it or not yeah yeah you maybe think of the analogy of like you tell an intern to do something you don't give them enough instructions yeah it turned out bad but you're gonna blame the intern precisely yeah it's a lot like that where they've gotten the past month or so they've gotten to crazy good levels and it's also because because they're so good in software in particular right like a lot of the process of software development was in terms of basically ensuring that nothing falls through the cracks and like double checking things and all that nature but a lot of people are trusting the code that's being built by this now because it's consistently that good but all it takes is a few times it being consistently not so like there is like you still need a person for sure but their role is more of like an ai checker than anything else now at this point a qualified ai checker us finalizing stable coin rules we talk about how the internet is messing with all of us all the time so if you're into that kind of thing definitely hit follow