hailey
banner
hailey.at
hailey
@hailey.at
engineering stuff, safety tools, and patterns
there’s an irony here too where many of the same folks will yell about the use of AI when AI actually is bringing about a ton of advancements in combatting the very abuse they say they are against. it doesn’t add up and is all vibes based.
January 10, 2026 at 7:05 PM
right now it’s trials in the sky remake with modified asset files to have joshua in the princess costume from the gender swap cutscene
January 10, 2026 at 6:55 PM
we had an internal verification rules document where item 18(C)(b)1 was “Jerry must never be verified”. sorry @jcsalterego.bsky.social
January 10, 2026 at 6:53 PM
my only regret is there are only so many weeks remaining of this glorious state
January 10, 2026 at 6:51 PM
👁️
January 10, 2026 at 6:04 PM
hiring is fucking toughhhhh rn…
January 10, 2026 at 6:04 PM
gremlin barry as always
January 10, 2026 at 4:53 AM
both ask it ahead of time to not do that (tell it to play the role of an interviewer) but also just ignore the answers it gives you - or at least take them constructively and figure out why that was the answer/feedback it gave. dont just take it at face value. try to find better ways than it gives u
January 10, 2026 at 4:51 AM
yea that wasnt cool at all. frankly pretty fucking disgusting
January 10, 2026 at 4:47 AM
i think its actually kind of cool that they might have been trained off of something that i wrote. like something about writing code to me is that it gets to affect people and products that i never intended for it to affect, and if an llm ever gave someone code based even 20% off my code, thats dope
January 10, 2026 at 4:46 AM
tbh i think i was just happy that no one took it to some weird degree and made a counter repo or something to try and prove and opposite point in a snarky way - a direction this whole thing absolutely could have gone
January 10, 2026 at 4:43 AM
don't let it give you answers. don't let it tell you what you should or shouldn't do. but absolutely do use it as a feedback provider. iterate off of the feedback it gives you until you feel like you understand why you agree with the feedback - both positive and negative - that it gives you
January 10, 2026 at 4:41 AM
recently enough went through the motions with one and it was a really great help in taking the scope of an abstract interview problem that was presented, laying out what the goals for that interview should be with a target company in mind, and then be a soundboard for the inputs i gave it
January 10, 2026 at 4:41 AM
oh no definitely, i think they made good points and admittedly even the prs that got made were still at least tasteful in language etc
January 10, 2026 at 4:38 AM
ya imo the whole going and tormenting the repo with PRs and stuff was....idk, a little bit edgy on "our" side's part, but im just glad the actual discussions were not super toxic and full of attacks the way they could have been
January 10, 2026 at 4:35 AM
i was pleasantly surprised with how tame the conversations went between people here for the most part. wasn't perfect but there was significantly less dunking and stuff that happened from a lot of folks and that as better to see than just outright bullying
January 10, 2026 at 4:26 AM
yea she deactivated it wasn’t a ban
January 10, 2026 at 2:45 AM
as a throughout test i had it running on the replies to about 20 accounts which were mostly the largest political and news organizations on the platform, and it was able to handle it fine
January 9, 2026 at 10:17 PM
yea i ran a labeler that used gemma running on a framework desktop that was doing some analysis of comments for “bad faith” observation. the results were fairly decent and well within the realm of acceptable for my own person use
January 9, 2026 at 10:17 PM
update: it was not good
January 9, 2026 at 8:53 PM
it is almost certainly not good but i’m going to choose to believe it’s good
January 9, 2026 at 8:45 PM
counterpoint
January 9, 2026 at 8:30 PM