On: Grieving the loss of details

[kaffekaka]

It is still an open question though exactly how big the need will be to propose good architecture or review slop. I petsonally think it will still be important, but it will vary by application for sure.

[abalashov → kaffekaka]

Take some comfort, if you like, in the notion that LLMs are limited to emitting convolutions of their training, plus whatever is stuffed into context.