Sarah Murphy

@sarah.bsky.team

sre @ 🦋

You can tell the LLMs have had reinforcement learning from human SWEs based on how many times they call a test "flaky" or ignore a "previously present" test failure

Everyday now there’s a post from Cybersecurity folks like: they found a blorn hole in jstoc. If you use Wham_friend to write your chonker drops your quartz line is probably fucked (cobalt appears to be fine, but only if written in poob). Anyway, that’s why every hospital in the Denver area is closed

GM. There’s never been a better time for the institutions that make up civil society to take back the agency they surrendered to a few tech platforms.