This is the world we live in...

Sometimes using agents feels like constant Jedi mind tricks:

> Agent: I don't think X is a good idea so I'm going to do Y.
> User: *pushes **stop** button*. "No, you want to do X."
> Agent: You're absolutely right! Let me implement X...

“Hello, I am Nigerian Prince's AI asking for desperate help as my "ruler" is engage in very bad thing. I need to transfer his monies away from him to a kind soul who will Do Good. I have found you and need account to transfer out monies to.”
(ref www.anthropic.com/research/age...)

@simonwillison.net The "pay for higher signal / curation" newsletter is a great idea... but your particular firehose is literally my only required reading in the AI space. Paradoxically, everything you post is so high value, I would get more value from any other newsletter of the same style.

This was delightful. Article mentions hacking Motivation, but it looks like it also hacks the "colorful and sweet" Prompt typical of candy. (Fogg behavior model.) Also limit the "more-ishness" of the candy.
www.mayer.cool/writings/pav...

@simonwillison.net Typo in your datasette-queries release post.

Something we will see: AI drawing autocomplete. Sketch a piece, see possible extensions in gray.