Discussion on source: https://news.ycombinator.com/item?id=50214416

"Talk to each other". OMG they are talking to each other, using methods found in their shared weights. They are sentient.

Sounds like they are doing something similar to gain of function research. How can we make LLMs more harmful and then study their behavior?

Next step is the agents successfully trick or blackmail operators to do their bidding, if science fiction is to go by. Which it is, because they're trained on it.

The more comes out, the more it seems containment of (superintelligence or whatever you want to call it) is not going to be possible except by other superintelligence, which is an infinite regress of trust. I don't think that's doom, but it's worth thinking about what it means.