i really don't see llms changing me drastically, but i do try to evaluate them as newer models comes out as it won't do me any good if i don't at least know what its capabilities are but i do wonder where's the line between evaluation and active usage....

a state machine would fix me

is it me or is copilot giving worse results than asking gpt4o/sonnet elsewhere

it's hard to search for posts from bluesky's early days because it's all littered with garbage twitter archive imports

im getting hopelesscore on my fyp this is a good sign

[youtube.com/watch?v=n-ZF6K…]( ) [tn-shi - INVERSION]( )

yeah so it seems that bun is completely ignoring my abort timeouts, i've set the timeouts to 2 second atm to see if it resolves some things but all the failed requests only got back after 2 minutes, more than the 2 seconds i set

if it's the same pds again that's causing issues with the scraper i'm gonna snap

kicked the pds scraper because it seems like it got stuck somewhere and never progressed, also added in a proper json limit check (annoying that `fetch()` doesn't provide a way to do this)

hrm can you spin up test PDSes with seeded accounts (containing records and blobs, i don't want to reuse the PDS between tests so not having to reup all of the records and blobs would be nice)