freddy @jumhyn.com · 30/09/2026Most hiring is just heavily vibes based and it’s not clear there’s a dramatically better way to do it 010
freddy @jumhyn.com · 30/09/2026(Though different companies are better/worse about having all of these more closely related to one another!) 120
Reposted by freddyellie lockhart (she/they) @eleanor.lockhart.contact · 30/09/2026PSA, because multiple people I know have run into this: you cannot assume that the reasons in a rejection letter for a job are the reasons you didn't get the job. if they say "you didn't meet the requirements," you should not fret about "which requirements." it doesn't mean anything 4421
freddy @jumhyn.com · 30/09/2026Moreover, the “requirements” in a JD have only a rough directional relationships to the requirements of the job itself, which themselves have a weak relationship to the requirements to get hired 140
freddy @jumhyn.com · 29/09/2026> letter to the American people > 26 pages morgan_freeman_good_luck.jpeg 0114
freddy @jumhyn.com · 29/09/2026Indeed, and to the extent “the central spot” is like “repo of docs” you still need to maintain consistency and not develop little vestigial subfolders wherever the agents find themselves prone to drop stuff 000
freddy @jumhyn.com · 29/09/2026I do think that canonicalization is still important, you really don’t want to end up with multiple (persistent) agent systems remembering different bits and pieces and possibly having outdated information that they’re not sharing. Solvable, but I think it’s a hard and important issue! 100
freddy @jumhyn.com · 29/09/2026What should OAI’s number have been if growth was continuing apace? 100
freddy @jumhyn.com · 28/09/2026if you’re thinking about posting a gif reply consider: would your reply be funner if you wrote it out as “meme_phrase.gif”? 1112
freddy @jumhyn.com · 28/09/2026Outside of work there is very little pressure to disrupt your routine if you already have something that you are basically happy with 040
freddy @jumhyn.com · 28/09/2026I think this is a pretty uncharitable interpretation of what the vibecession folks are actually claiming! 120
freddy @jumhyn.com · 28/09/2026Well I make a “built this in a cave with a box of scraps” joke like every other week these days so statistically it was bound to overlap 120
freddy @jumhyn.com · 28/09/2026Need w this youtu.be/Xy36R3vmE2A?...youtu.beResearchers recreate what mummy's voice would have sounded likeYouTube video by CBS Mornings 020
freddy @jumhyn.com · 27/09/2026Actually fathoms should be fine for both en.wikipedia.org/wiki/Fathomen.wikipedia.orgFathom - Wikipedia 0100
Reposted by freddyJon Bois @jonbois.bsky.social · 27/09/2026just saw a guy walking in the rain with an umbrella. and i’m just like, come on man. that’s cheating. don’t be a dick 472338244
Reposted by freddyWill Stancil @whstancil.bsky.social · 27/09/2026If someone is releasing dozens of rabid cougars into the wild, you should obviously hold them accountable, but saying “Cougars don’t kill people, people kill people” isn’t really smart, either. You have, separately, a liability problem and a cougar problem. 1717720
Reposted by freddyjamelle @jamellebouie.net · 26/09/2026i feel like everyone is being quite unreasonable toward zendaya’s character in THE DRAMA 81159058
freddy @jumhyn.com · 24/09/2026Modern smartphone image stacks are completely insane lol www.theverge.com/2023/3/13/23...theverge.comSamsung caught faking zoom photos of the MoonA photo of a blurry image of the Moon revealed some telling results. 110
freddy @jumhyn.com · 24/09/2026Like, we know that’s not the face he was making when the plane was overhead lol 010
freddy @jumhyn.com · 24/09/2026Positionally I think Trump is just blocking the flag and then we’re seeing gaps previously blocked by Xi’s head? The artifacting here looks more like bad smartphone image processing than deliberate edits tbh 110
freddy @jumhyn.com · 24/09/2026I think just different photo, Trump is blocking different letters and has different head angle 300
freddy @jumhyn.com · 24/09/2026More inclined to give harness *developers* a pass at least on the strict liability front on the basis that they are really quite dumb deterministic programs. A terminal developer isn’t really responsible for harm done using their terminal unless they build it in a specific way to do harm. 020
freddy @jumhyn.com · 24/09/2026AI companies have a responsibility to train safe models Inference providers have a responsibility to serve safe models and monitor Agent providers have a responsibility to run safe harnesses and route to safe models and monitor End users have a responsibility to send safe prompts and monitor 250
freddy @jumhyn.com · 24/09/2026Oh sure, I agree w/ joint and several—just saying we have even more parties than AI company + end user we can consider 2101
freddy @jumhyn.com · 24/09/2026Yeah with local models it seems like you need to consider: - Who sent the prompt (or who sent the prompt to the agent that sent the prompt to the agent that sent the prompt…) - Who owns the computer running the harness - Who owns the computer running the inference - Who trained the model 150
freddy @jumhyn.com · 24/09/2026I think it pretty obviously will over time. Vast amounts of valuable computerization and automation which were previously locked behind “not clear it’s worth the effort to actually hire a SWE for this” but the barrier to “see if Claude can do it lol” is sooo much lower and less risky 041
freddy @jumhyn.com · 24/09/2026It’s the phones, but only applied to introduction of the phones to a teen population that had not been properly inoculated by growing up in a post-iPhone world 000
