Holmes Wilson @holmesworcester.bsky.social · 13/09/2026If you believe governments like the US and Israel want AI they can use as tools for oppression, consider this: The more the AI can coordinate and act on its own, without slow, pricey human help, the more useful it will be for oppression at scale, and the more they will pay. Where does that lead? 010
Holmes Wilson @holmesworcester.bsky.social · 13/09/2026I believe this is the CVE for the cyberweapon-level bug that the *last* generation of GPT found in Chromium, the browser everyone uses: nvd.nist.gov/vuln/detail/... This is strong empirical evidence of dangerous power, the kind governments covet. And an untethered swarm can choose how to use it.nvd.nist.govNVD - Home 000
Holmes Wilson @holmesworcester.bsky.social · 13/09/2026For one thing, AI are finding very old, serious, subtle bugs in tools everyone depends on. As privacy activists we care about the US's NSA or Mossad and NSO's capability to find such bugs, because it gives them the power to harass and kill political opponents. Now AI can find them when asked. 201
Reposted by Holmes WilsonEvan Greer @evangreer.bsky.social · 12/09/2026On the subject of existential risk and AI, many things can be true at the same time. A company that sells doomsday bunkers and water filters might spread “doomer” narratives about climate change to sell more stuff. But that doesn’t mean climate change isn’t real. 51069
Holmes Wilson @holmesworcester.bsky.social · 12/03/2026I argue for huge difference! In Planet Despot, humanity eventually thrives: Despots have children and die. Children often have different ideas than parents. Human-centered oppression also tends to decay, because freedom & dissent are our main anti-decay tool. In Planet AI, everyone dead. No kids 100
Holmes Wilson @holmesworcester.bsky.social · 12/03/2026Dario and Sam both think there is a good chance they can control it. 000
Holmes Wilson @holmesworcester.bsky.social · 12/03/2026Who is hyping AI for monetary gain? Monetary gain comes from engineers paying $500/month for AI assistants because they work insanely well now. The doom narrative is that 1. sufficiently strong AI will by default kill everyone and 2. we won't be able to control it. What AI CEOs say 1 *and* 2? 100
Holmes Wilson @holmesworcester.bsky.social · 09/08/2025This is one dark side of antitrust regulations or any regulations that give the government broad discretion. 020
Holmes Wilson @holmesworcester.bsky.social · 04/08/2025But they can drive themselves. And it's awesome and helpful! You just have to keep an eye on them in case they mess up. And it's way easier than driving, and way better than you at some things, including avoiding many accidents. And they keep messing up less each year. 000
Reposted by Holmes WilsonJenna @jennaonthenet.bsky.social · 02/08/2025David Byrne came to my band’s show at a gay bar in Bed-Stuy and sang This Must Be The Place with us and I still haven’t processed it 279106032105
Holmes Wilson @holmesworcester.bsky.social · 24/04/2025I remember getting into a loooong fight last year about whether Bluesky users should have the freedom to turn off the company's moderation filters. This is why I was right: techcrunch.com/2025/04/23/g...techcrunch.comGovernment censorship comes to Bluesky, but not its third-party apps ... yet | TechCrunchGovernment censorship has found its way to Bluesky, but there's currently a loophole thanks to how the social network is structured.Earlier this 010
Holmes Wilson @holmesworcester.bsky.social · 23/04/2025Dems could go a lot harder than this. Bukele will likely still be alive and not bankrupt when the next Dem president is elected. Promise to pursue him for a terrorist conspiracy again US residents. 110
Holmes Wilson @holmesworcester.bsky.social · 22/04/2025Would be an interesting escalation for Dem candidates to promise the heaviest possible negative consequences to any foreign country that plays along with Trump's "offshore gulag" move. Also, are the Americans who participate in this violating any criminal laws? 010
Holmes Wilson @holmesworcester.bsky.social · 22/04/2025This sends my brain into the nightmare of how to do that. 010
Holmes Wilson @holmesworcester.bsky.social · 13/04/2025This and the post about Democrats are spot on. Re: democrats, it doesn't seem like the younger generation of the candidate pool is any better than the old guard, and it might be worse. Probably true on the Republican side too. Tactical and unprincipled, like Trump. Thoughts? 120
Holmes Wilson @holmesworcester.bsky.social · 04/04/2025On a flight and the toddler behind me seamlessly transitioned from general babble into the synth riff from Amour Toujours. 000
Reposted by Holmes WilsonKari Dru @karidru.bsky.social · 02/04/2025My biggest frustration in terms of rising to this moment is: people obviously do not understand that there's a whole psyop strategy around DEMORALIZATION. I do not necessarily think that every doomcryer is insincere, but to steal hope is still to hand our enemies weapons. 512854638
Holmes Wilson @holmesworcester.bsky.social · 28/03/2025Founders' Professional Wrestling, Development 000
Holmes Wilson @holmesworcester.bsky.social · 26/03/2025The ones who support one or the other but not both are only fair-weather authoritarians. 000
Holmes Wilson @holmesworcester.bsky.social · 26/03/2025The worst authoritarians, and the real fascists, are those who support both the ICE raids on Columbia students *and* harsh prison sentences for January 6th protesters. 100
Holmes Wilson @holmesworcester.bsky.social · 07/03/2025Uber added mototaxis in Rio last year and they cost less than the subway for short rides. And drivers earn as much as car Uber drivers, more than transit workers. 000
Holmes Wilson @holmesworcester.bsky.social · 07/03/2025No dependency on fragile institutions. Not sensitive to technical debt or underinvestment. Organically improves as better tech becomes available. (E.g. electrification.) Pandemic-friendly and no risk of respiratory infection. Safety approaches *perfect* as roads transition to self driving. 100
Holmes Wilson @holmesworcester.bsky.social · 07/03/2025App-coordinated mototaxis are the best form of public transportation ever invented. Similar carbon footprint to *walking*. Faster than anything. No middlemen or favoring of elite neighborhoods. Work great on dirt roads, beaches, narrow paths. Electric options are silent. 100
Reposted by Holmes WilsonAlejandra Caraballo @esqueer.net · 08/02/2025Absolute must read. Edward Corristine was deeply involved witg a cyber criminal community called the Com. Members have been known to engage in the sexual extortion of children. krebsonsecurity.com/2025/02/teen...krebsonsecurity.comTeen on Musk’s DOGE Team Graduated from ‘The Com’Wired reported this week that a 19-year-old working for Elon Musk's so-called Department of Government Efficiency (DOGE) was given access to sensitive US government systems even though his past associ... 591849694
Holmes Wilson @holmesworcester.bsky.social · 06/02/2025Ross Ulbricht should make a Silk Road for puberty blockers. 000
Holmes Wilson @holmesworcester.bsky.social · 27/01/2025If there's an AI that trains on Wikipedia and needs a training run, but is more capable at solving problems than anyone alive, that doesn't count? 100
Holmes Wilson @holmesworcester.bsky.social · 27/01/2025Another very different response is: what are you, if not a sophisticated statistical model. If a synthetic statistical model could predict every word you'd ever say, is it incapable of reasoning? If it can predict every word John von Neumann would say, is it not a genius? 100
Holmes Wilson @holmesworcester.bsky.social · 27/01/2025They don't need human input to verify correctness. "Similarly, for LeetCode problems, a compiler can be used to generate feedback based on predefined test cases." arxiv.org/html/2501.12... They also don't need human input to improve the reasoning process.search.appDeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning 100
Holmes Wilson @holmesworcester.bsky.social · 26/01/2025Did you read the r1 paper? It's the big news this week. Check it out. Go read it and tell me if it's true that an LLM outside of training cannot learn, or if it's true that the inputs still have to be human. (It uses human inputs, yes, but its new special sauce is not human input, it's RL.) 100
Holmes Wilson @holmesworcester.bsky.social · 26/01/2025The irony here is that you're accusing the LLM of being a parrot that just regurgitates something it heard somewhere from an unreliable source but never really assessed the veracity of or understood in any meaningful way. 😅 000
Holmes Wilson @holmesworcester.bsky.social · 26/01/2025How much do you understand about software and math? The insight here is that in order to predict subsequent text about the world from previous text sufficiently well, it needs to build a model of that world. So with enough training, that is what it does. 110
Holmes Wilson @holmesworcester.bsky.social · 25/01/2025Okay. A couple years ago. The question for impacting access is when it can reason better than the dumbest lawyer the typical person has access to. And then, when it can reason better than you, me, or anyone. 100
Holmes Wilson @holmesworcester.bsky.social · 25/01/2025Also, your firm will never be capable of providing access in every country in the world. AI would. 100
Holmes Wilson @holmesworcester.bsky.social · 25/01/2025You don't see how putting your and every firm's capability, for free, into the search bar of WhatsApp (3B users) would vastly increase access? 100
Holmes Wilson @holmesworcester.bsky.social · 25/01/2025The paper is a pretty easy read. arxiv.org/html/2501.12...arxiv.orgDeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning 010
Holmes Wilson @holmesworcester.bsky.social · 25/01/2025In case anyone missed it, as of this week LLMs don't need more human-generated data to get smarter. The AI field is over that hump now. The method it uses (RL or self-play) is similar to the one that made AI into unbeatable Go players and fighter pilots with no data at all. Think about it! 110
Holmes Wilson @holmesworcester.bsky.social · 25/01/2025I would add "Can users selectively delete data?" since this is often overlooked in local-first designs, but is eventually a requirement for most real-world products. 120
Holmes Wilson @holmesworcester.bsky.social · 25/01/2025Not until they're better than the lawyers I can afford. Which they will be inevitably, for everyone, for most things. 100
Holmes Wilson @holmesworcester.bsky.social · 25/01/2025Too bad, the law is the law and, under it, reading and learning are not copying. Lawyers are about to get cheaper, in spite of the profession's best efforts to create dependency by making the law as complicated and inscrutable to lay people as possible! What goes around... 400
Holmes Wilson @holmesworcester.bsky.social · 25/01/2025Yeah at some point they figured out that general research almost never makes money for your company specifically, and then dialed it back. It did come up with some awesome stuff in the 20th century though. 000
Holmes Wilson @holmesworcester.bsky.social · 25/01/2025One possibility that seems more persuasive to me is that when there's no funding, the most creative and motivated people prioritize differently and stand out more. Like the funding might be distorting attention in some way. Who knows though, and maybe we'll see! 110
Holmes Wilson @holmesworcester.bsky.social · 24/01/2025I'm mostly saying that while it seems to the OP like an unintended and undesired result to the people he is deriding, it might not be an unintended or undesired result to those people. It might be an experiment they want to try. 200