[chattable]

https://www.reddit.com/r/ControlProblem/comments/13j8nva/examples_of_ai_safety_progress_yoshua_bengio/?utm_source=share&utm_medium=web2x&context=3
the all text ai agents should not be banned because they cannot do much . they only understand how to use text .
if the agi can only think in text it would be a great way to monitor it's thoughts. if you could keep it using text to think when developing into agi .
that would be the way to go to solve the control problem.

[WriterOfMinds]

Banning AI agents doesn't seem feasible, whether it would be a good idea or not. This isn't nuclear power ... it's very easy to hide.

They might be able to crack down on the big model-training operations, but more computationally efficient work would pass under the radar.

[MikeB]

Who decides what is/isn't AGI?

Does a completely white-box, believable character in a video game, need all their lines vetted by someone...