It’s starting…
Channel: Matthew Berman
Uploaded:
Duration: PT2692S
Read a transcript preview of “It’s starting…” from Matthew Berman, uploaded on 2026-06-05.
Transcript preview
Language: English (original, YouTube automatic captions)
[0:00] AI is now literally building itself. And
[0:03] according to Anthropic, that means two
[0:06] things. One, society is not prepared.
[0:09] And two, we should actually slow down
[0:11] development. That is incredibly
[0:13] self-serving by Anthropic, but let me
[0:16] explain it all. And if you like me
[0:18] breaking down things like this, please
[0:20] subscribe to the channel, like the
[0:22] video. It really does help us grow. And
[0:25] let's get on to the paper. When AI
[0:28] builds itself, our progress toward
[0:31] recursive self-improvement and its
[0:33] implications. So, it starts out just
[0:36] straight to the point. We are delegating
[0:38] a growing share of AI development to AI
[0:42] systems. Given enough compute, that
[0:44] trend points to an AI system capable of
[0:48] fully autonomously designing and
[0:50] developing its own successor. Now,
[0:52] interestingly enough, they say recursive
[0:54] self-improvement is not inevitable. It
[0:57] definitely seems inevitable at this
[0:59] point, but there is one missing
[1:01] ingredient that we don't have today. And
[1:03] so, stick around because I'm going to
[1:06] explain what that is later in this
[1:07] video. And of course, they have to
[1:09] mention safety because they're
[1:10] Anthropic. Full recursive
[1:12] self-improvement also might increase the
[1:14] risk of humans losing control over AI
[1:16] systems. That's putting it lightly. I
[1:18] definitely agree. This is a major
[1:21] alignment problem that needs to be
[1:23] solved. All right. Now, what they're
[1:25] going to show in these little graphics
[1:27] that they put together, which by the
[1:28] way, beautiful. Just top-tier graphics
[1:31] on the Anthropic team. They are going to
[1:34] show the progression of h