Luis Villa @lu.is · 1hRight, to do that you’d have to write a lot of specs and a lot of tests. Which is doable but a lot easier when you’re building them organically alongside a small code base. 000
Luis Villa @lu.is · 2hAdmittedly the only places I’ve ever seen elephants in the street were NYC and Coral Gables Florida, but I stayed mostly in bigger cities in India :) 110
Luis Villa @lu.is · 3hNo cows, just mountain lions www.sfgate.com/bayarea/arti...sfgate.com'Majestic' mountain lion captured after hourslong SF standoffThe young mountain lion was tranquilized before being removed from the area. 110
Luis Villa @lu.is · 3hThe 2nd project’s code will almost certainly be better. Even if all 400 of the LLM PRs are rejected, building the testing and analysis flow to handle that will mean better code. But that’s going to be a hell of a hill to climb for existing projects (and their under-appreciated, unpaid maintainers). 070
Luis Villa @lu.is · 3hEg I saw this week a FOSS project that had received 400s+ LLM-filed bugs over the weekend. Existing FOSS projects will rightly say “fuck that”. But I’m already seeing projects that will just feed that to their own LLMs and CI processes for validation (or rejection) before humans waste a second. 130
Luis Villa @lu.is · 3hHuman drivers do it all the time to indicate in busy crosswalks that “yo, I gotta get across eventually, let me through”. Or when they are on their phone and not paying attention, or just irritated that pedestrians exist at all. Waymo only does the first of these. 110
Luis Villa @lu.is · 3hThe Waymos in S.F. are both reading cues and cueing themselves, and have been for about a year. But it is problematic (many “driver” cues are literal violations of the law, like nudging into the crosswalk) and no idea how they’re thinking about working across car cultures. 120
Luis Villa @lu.is · 3hThis is why I think on net, going forward, LLMs are a huge win for software quality. But the backlog is immense and lots of our supply chain will (rightfully) say “fuck it” and walk away. 2174
Luis Villa @lu.is · 4hdon’t discount the 15 year old’s feeling when he has noticed this is a giant penis, or the mid-life man who has also noticed it is a giant penis 030
Luis Villa @lu.is · 4hA friend worked on reverse-engineering the Apollo heat shields for the new capsules. Lots of the learning boiled down to “whoa we can’t manufacture that anymore because it, and it’s manufacturing process, was mega-toxic to all forms of life everywhere“ 090
Reposted by Luis VillaErika Hall @erikahall.bsky.social · 5h1938 ad for cardboard boxes goes way harder than it needs to 77414
Luis Villa @lu.is · 8hthe next thing he does like a normal person will be the first thing he does like a normal person. (Common superstar problem, to be fair!) 120
Luis Villa @lu.is · 9hJudges in particular. Why ask for a tough sentence when the judge drove to work, almost certainly speeding the whole way, today? (Same problem, sad to say, as sentencing and “boys will be boys, why ruin their lives” rape culture.) 040
Luis Villa @lu.is · 9hen.wikipedia.org/wiki/Motonor...en.wikipedia.orgMotonormativity - Wikipedia 120
Luis Villa @lu.is · 9hMeta learning from “other companies“ failed hardware rollouts Which… other companies, Mark 130
Luis Villa @lu.is · 9hAlso if I’m reading it correctly it’s actually revoke after the fifth and only if the fifth hurts someone 220
Luis Villa @lu.is · 9hTaps yet again the “so far society has a vastly better track record of holding autonomous vehicle companies responsible for injury and death than holding human drivers responsible” sign 061
Luis Villa @lu.is · 9hIt is both great that these passed and utterly damning that this is the best we could get. The last two in particular in any sane society would be ”never drive again” rather than debating how many points on a license or how long the suspension is. 6316
Luis Villa @lu.is · 10h(Most companies probably shouldn’t either! But that‘s a more complicated conversation…) 010
Luis Villa @lu.is · 10hThis research is important not just for the copyright nerds. It’s a great example of informed auditing to help society understand the implications of LLMs. Bluntly, governments should never use models that scholars like @afedercooper.bsky.social can’t audit. 1195
Luis Villa @lu.is · 12hi loved mine and when I broke it and wanted to buy a replacement the website was full MAHA. 010
Luis Villa @lu.is · 23hAs a Duke '01 alum, I have long wondered how it was that Duke ended up without frat houses when every other Southern campus has them. (And thank you for your contemporary service.) 010
Luis Villa @lu.is · 30/09/2026Need a version of this diagram but to explain all this. (Joking but… not joking?) commons.wikimedia.org/wiki/File:Br...commons.wikimedia.org 130
Luis Villa @lu.is · 30/09/2026They must certainly have discussed that, and I'm sure they're gathering data via their deployment in Florida now. 120
Luis Villa @lu.is · 30/09/2026silver lining, maybe having a Korematsu will make it easier to pack the court? 100
Luis Villa @lu.is · 30/09/2026Body and car count of just news from @sfchronicle.com and @missionlocal.org just this afternoon, just in SF and the peninsula, just those reported: three dead, 12 or 13 cars trashed. We need a plan to ban human drivers, which means we need a plan for transit, bikes, and yes, AVs. 040
Luis Villa @lu.is · 30/09/2026And it isn't *very* visible in public, but you can see it in the less-than-confident responses to questions about "do you really need to inch into the crosswalk" as opposed to the still very-confident messaging from them on safety generally. 130
Luis Villa @lu.is · 30/09/2026You can now also see the reverse: Waymo has realized "oh, maybe strict adherence to every traffic law is not very workable" which means suddenly the philosophical questions are waaaay more complicated. 1110
Luis Villa @lu.is · 30/09/2026I like this framing of LLMs as "intelligence amplification". For Wikimedians and Wikipedia, the question is where are humans involved: still subject to the healthiest and best of our community constraints? Or does that migrate fully to big model providers? backofmind.substack.com/p/because-i-...backofmind.substack.combecause I say so, that’s whya loop cannot close itself 000
Luis Villa @lu.is · 29/09/2026There's a lot of the same deep puzzle-solving motivation for AV people, but (1) the hard physical constraints tend to ground people, so that it's more "passionate vocation" and less "religion"; (2) at Tesla in particular, lots of those deeply curious people have fled due to Elon being Elon. 090
Luis Villa @lu.is · 29/09/2026We were promised Lex Luthor and Dr. No and instead we got variations on every idiot in Dr. Strangelove. 050
Luis Villa @lu.is · 29/09/2026But I think that core fact fundamentally means the discussions start in different places. 1120
Luis Villa @lu.is · 29/09/2026AV also has more concrete notions of safety: are people dying? The LLM safetyish discussion operates largely in fantasylands of future hypothetical harms. This is oversimplified; obviously the FSD people don't give a shit about body count and some AI safety people discuss present-tense IRL harms. 1321
Luis Villa @lu.is · 29/09/2026Somewhere, someone at Duke is kicking themselves that they didn't do the same when the Cameron troughs went away 000
Luis Villa @lu.is · 29/09/2026Too bad. We’d be better off if aggregation theory were taught in antitrust and to every incoming congressperson, but that doesn’t excuse brain rot. 010
Luis Villa @lu.is · 29/09/2026It wasn’t so much his personal 9/11 as it was vice-signaling to his oligarch buddies. They wouldn’t shut up about it, because they think all their employees are stupid and woke, and dislike any binding constraints like “maybe you shouldn’t be racist. He was echoing them. 1201
Luis Villa @lu.is · 29/09/2026I suspect that best practice separates identity from keys, since if someone blocks a key you don’t want the entire identity blocked? But maybe that’s the low-risk, low-effort option? 010
Luis Villa @lu.is · 29/09/2026Or rather, for anonymous APIs. Many of the best open properties can be had from keyed APIs. (But we don’t have much of a coherent theory of which/how). 171
Luis Villa @lu.is · 29/09/2026Been very sad to see him suffer from advanced pundit brain combined with a bad case of In The Broligarch Group Chats. (I canceled my subscription right around your older example.) 080
Reposted by Luis VillaEd @ed3d.net · 29/09/2026you can stop watching this is pretty much all the "hey, this is your platform lock-in" stuff that folks have been worried about since token prices started crashing. the good news is, with tokens you can build ones that you control.openai.comDevDay 2026 RecapExplore more than 20 announcements from OpenAI DevDay 2026, including GPT-6 Astra, ChatGPT, Codex, APIs, security, and new tools for builders. 3403
Luis Villa @lu.is · 29/09/2026It will if the remigration nutters get their way, but yes, they'll have to be in charge for a long time to actually have Japan-level impact. 010
Luis Villa @lu.is · 29/09/2026the track record of boring ol un-revolutionary politics is, to be fair, not great in the US of late! But the track record of revolutionary politics in the US is basically a graveyard of noble ideas since the IWW more or less. 1110
Luis Villa @lu.is · 29/09/2026That actually also defines the US at the current moment, so I think you left out an important “thinks things can be improved”. But yes, Japan. 110
Luis Villa @lu.is · 29/09/2026All the carbon and pollution concerns boil down to “so you should work like hell to elect a party that cares about those issues; demand they fund and staff enforcement agencies competently; and demand streamlining and prioritization of enforcement.” But that’s hard to get angry about online. 1292