My reflection more generally is: LLMs are nondeterministic and thus not deeply reliable, but sitting and drilling with one model, on autonomy in the context of a specific set of tools and use cases, can pay huge dividends.

This is not an area in which I am a subject matter expert. However, this doesn't seem out of the realm of possibility, and dovetails with other expert concerns.

And a quick crop on one of those for detail

if you're ever Photoshopping your own face into the smoke of the Twin Towers, Adobe should have a pop-up that says "are you sure that whatever it is you're doing is a good idea"

A lesson from depression: Take time to intentionally celebrate wins, regardless of their size.

Train your brain that there is joy to be had amidst the fear and anger and grief, and you want to attend to the joy.

However - as Brian Krebs has shown - even freezing your credit can leave you vulnerable, as credit agencies really aren’t interested in protecting you.

BTW if you’re American and getting into threat research, freeze your credit with all three agencies now.

(But also decent advice for everyone, given the state of things.)