Not a Spring Onion on Nostr: If I may, and this could very well into a side topic: I think an interface *shapes* ...

If I may, and this could very well into a side topic:

I think an interface *shapes* the control the human has.

In the best case, it allows the human to do what that human wants, while introducing some additional information or friction where it makes sense.

In the worst case, it constantly gaslights the human, presenting choice where there aren't any, simulating control while actually preventing it.

An interface can be propaganda.

And more back to the original post:

LLMs can be and are their own form for gaslighting/propaganda.

"Oh, sorry. I made a mistake."

The LLM did not actually make a mistake. It is just following a script (conversation script, not code) to appease the human and muddy the actual problem.

When vibe coders complain that the talking parrot just deleted their whole code base, that is where the facade of an AI "coworker" breaks and the truth of a fancy random number generator appears.