Dean W. Ball @deanwb.bsky.social · 27/03/2025t.co/HoraJ0o77Lt.cohttps://www.hyperdimensional.co/p/where-we-are-headed 010
Dean W. Ball @deanwb.bsky.social · 27/03/2025There are tons of technocratic AI policy proposals, but little concrete intuitions about the near-term future to motivate those proposals. The result is that the field sometimes feels disembodied, devoid of a concrete vision. Here is what I think is coming. 492
Reposted by Dean W. BallNathan Lambert @natolambert.bsky.social · 12/02/2025OpenAI releasing a model spec is one of my favorite things about them. All frontier labs should be doing this. Closes the gap between intentions and results, so you can interpret them better. openai.com/index/sharin...openai.comSharing the latest Model SpecWe’ve made updates to the Model Spec based on external feedback and our continued research in shaping desired model behavior. 0214
Dean W. Ball @deanwb.bsky.social · 16/01/2025Enjoy! t.co/eQczeWf4Vlt.cohttps://www.hyperdimensional.co/p/diffusion-confusion 040
Dean W. Ball @deanwb.bsky.social · 16/01/2025My latest, on the AI export controls generally and the diffusion rule in particular. This regulatory adventure, whatever you think of it on the merits, is a case study in how rules beget more rules—how regulation grows, like a plant or a weed. 1111
Reposted by Dean W. BallTimothy B. Lee @binarybits.bsky.social · 14/01/2025Excited to launch a new podcast with @deanwb.bsky.social about AI. It's called AI summer, and can be found on all major podcast apps. The first episode is an interview with political scientist @jonaskonas.bsky.social about AI policy in the Trump era. www.aisummer.org/p/jon-askona... 0195
Dean W. Ball @deanwb.bsky.social · 12/01/2025A little late, but--my 2025 'lookahead' post is here, covering some of the major AI products, research, and policy I'll be watching this year. I hope you enjoy! t.co/CZdR8zjsFH 080
Reposted by Dean W. BallTech Policy Press @techpolicypress.bsky.social · 09/01/2025American AI policy in 2025 will almost certainly be dominated, yet again, by state legislative proposals rather than federal government proposals, writes Dean Ball.buff.lyThe Coming Year of AI Regulation in the States | TechPolicy.PressAmerican AI policy in 2025 will almost certainly be dominated, yet again, by state legislative proposals rather than federal government proposals. 167
Dean W. Ball @deanwb.bsky.social · 09/01/2025Hey, it’s worked well for my specific uses (which often involve broad surveys of state policies), but I have heard lots of people saying it is mid for them. Ymmv! 110
Dean W. Ball @deanwb.bsky.social · 05/01/2025I think it mostly wouldn’t apply to the dod specifically but it absolutely would apply to any dod contractors with a presence in Texas. 030
Dean W. Ball @deanwb.bsky.social · 29/12/2024I would say I’ve heard these points made by people from many of the most prominent think tanks in dc. Would also point you to articles like this recent one from the Wall Street journal www.wsj.com/opinion/ai-c...wsj.comOpinion | AI Can’t Teach AI New TricksA lack of human prose leads to what’s called ‘model collapse’ and pure gibberish. 030
Dean W. Ball @deanwb.bsky.social · 19/12/2024Link here: www.hyperdimensional.co/p/one-down-m...hyperdimensional.coOne Down, Many To GoReflections on year one of Hyperdimensional 030
Dean W. Ball @deanwb.bsky.social · 19/12/2024Hyperdimensional is almost a year old! So I wrote a reflection on the first year, covering the origins of the project, SB 1047, where I erred, the impossibility of controlling technocracy, where I see the project going next, and more. Enjoy! 350
Dean W. Ball @deanwb.bsky.social · 11/12/2024I really wish more people criticized my own work, too. 040
Dean W. Ball @deanwb.bsky.social · 07/12/2024“Since we necessarily underestimate our creativity, it is desirable that we underestimate to a roughly similar extent the difficulties of the tasks we face, so as to be tricked… into undertaking tasks which we can, but otherwise would not dare, tackle.” -Albert Hirschman 050
Dean W. Ball @deanwb.bsky.social · 06/12/2024Heres the post: www.hyperdimensional.co/p/impact-ass...t.cohttps://www.hyperdimensional.co/p/impact-assessments-are-the-wrong 000
Dean W. Ball @deanwb.bsky.social · 06/12/2024Everyone I know who follows AI policy closely knows about these bills and is concerned. But there is no public outcry, like there was with SB 1047. It’s time to change that. 100
Dean W. Ball @deanwb.bsky.social · 06/12/2024One is already law (going into effect in 2026) in CO. Another will probably be law soon enough, via bureaucrat “rule making,” in CA. It has a good chance of passing next year in TX, VA, and CT. We’ll see it elsewhere, too. 100
Dean W. Ball @deanwb.bsky.social · 06/12/2024I’m referring to a slate of civil rights-based bills we will see throughout America. They’re among the most complex AI bills with a serious chance of becoming law. 100
Dean W. Ball @deanwb.bsky.social · 06/12/2024America is charging into an AI policy regime that will slow down adoption of AI, shrink model developers’ addressable market, and could deter frontier AI investment. It’s one of the worst ways to regulate AI, it could become a national standard soon, and it’s under-discussed. 110
Dean W. Ball @deanwb.bsky.social · 05/12/2024If chatbots stay in their current role—assistants I use for random tasks throughout the day—then ads have a negligible effect on their utility. If however chatbots become something like what the ai companies all aspire to—genuine “virtual chiefs of staff”—ads may be ruinous. 220
Reposted by Dean W. BallChris Painter @chris.bsky.social · 02/12/2024The world we live in today is already incredibly Cyberpunk, by comparison with the world of the 90s and early 2000s that I grew up in 191
Dean W. Ball @deanwb.bsky.social · 29/11/2024Evidently, yes. There is one item from the bulleted list that we omit in that passage (discrimination). 100
Dean W. Ball @deanwb.bsky.social · 28/11/2024The post is just that: a post summarizing my own views. And my view is that the focus on nebulous risks in the CoP, in the AI Act itself, and in various American policy documents is a foundational error. Also, I’ll just note that you are wrong: we don’t say every listed risk is easily measurable. 110
Dean W. Ball @deanwb.bsky.social · 28/11/2024I would suggest reading the articles you’re criticizing before you criticize them, in general. Neither miles nor i strawmanned anyone. if you think the current document is even close to well scoped, that’s fine! We just disagree. 100
Dean W. Ball @deanwb.bsky.social · 28/11/2024I also do not think it’s right to say that the government (the people who can fine these companies billions) should “merely” tell developers to “consider” what amounts to a massive range of nebulous risks. There is nothing “mere” about that! 100
Dean W. Ball @deanwb.bsky.social · 28/11/2024The portions we quoted and criticized are quoted directly from the piece. You can disagree with our criticism and believe these risks should be considered, but it is not a strawman to quote directly from a piece. 100
Dean W. Ball @deanwb.bsky.social · 27/11/2024Full post here: www.hyperdimensional.co/p/on-the-eu-...hyperdimensional.coOn the EU AI Code of PracticeA co-authored comment 000
Dean W. Ball @deanwb.bsky.social · 27/11/2024I was happy that I could find lots of common ground with @milesbrundage.bsky.social in our assessments of this draft--and it was a pleasure to work with him. 140
Dean W. Ball @deanwb.bsky.social · 27/11/2024One of our main takeaways is: focus on measurable, tightly defined risks, rather than "risks to society as a whole," as the draft Code of Practice currently does. This problem also affects US AI documents like the NIST RMF. It's no way to make policy, in my view. 320
Dean W. Ball @deanwb.bsky.social · 27/11/2024Today in our respective newsletters, @milesbrundage.bsky.social and I co-author a comment on the EU General-Purpose AI Code of Practice. We try to offer constructive comments as to how this document--which will likely be a standard for frontier AI dev compliance with the AI Act--can be improved. 4567
Dean W. Ball @deanwb.bsky.social · 26/11/2024The eu ai act should quite obviously be repealed. Unsalvageable. 010
Dean W. Ball @deanwb.bsky.social · 26/11/2024The EU General-Purpose AI Code of Practice—a document that almost exclusively regulates American companies—came out in draft on November 14. The deadline to submit comments is two weeks later, November 28, aka “Thanksgiving.” You have to admire the passive aggression. 020
Dean W. Ball @deanwb.bsky.social · 25/11/2024almost every proposed ai regulation should come with a sunset provision. 010
Dean W. Ball @deanwb.bsky.social · 24/11/2024autonomously self-replicating ai compliance documents 041
Dean W. Ball @deanwb.bsky.social · 24/11/2024In other words, do you have to write an algorithmic impact assessment about using AI to write an algorithmic impact assessment? 120
Dean W. Ball @deanwb.bsky.social · 24/11/2024If you use an llm to write an algorithmic impact assessment/risk management plan about your use of ai for “consequential decisions,” are those LLM-written compliance documents themselves about consequential decisions, thus requiring an additional algorithmic impact assessment? 120
Dean W. Ball @deanwb.bsky.social · 22/11/2024You might ask, reasonably, “what could a manuscript about agricultural epistemics in the antebellum south possibly have to do with ai?” But give it a try! 010
Dean W. Ball @deanwb.bsky.social · 22/11/2024As a general matter I have not enjoyed Kendrick’s work post-gkmc. I feel he somehow lost the sense of urgency that characterizes all great music. This new album gnx however is quite good so far (five tracks in) 000
Dean W. Ball @deanwb.bsky.social · 22/11/2024awesome book, recommended, especially for people thinking about ai 062
Dean W. Ball @deanwb.bsky.social · 22/11/2024Question: can one share links here without as much fear of being algorithmically disfavored? A test. youtu.be/FwPhiWZN6jQ?... (Really cool piece, especially love the piano at about 8 minutes)youtu.beCharles Mingus Sextet 1970 - Pithecanthropus Erectus (live video)YouTube video by Bob Juke 120
Reposted by Dean W. BallNathan Lambert @natolambert.bsky.social · 20/11/2024I love seeing RL continue its magnificent takeover that I have doubted so many times over. I can go back to being an RL person, I don't even need to masquerade as an "RLHF" or "NLP" person anymore. 1391
Dean W. Ball @deanwb.bsky.social · 20/11/2024The latter I believe. And possibly not even shift the OW to the manhattan project idea, but instead away from open source. 110
Reposted by Dean W. BallSeth Lazar @sethlazar.org · 20/11/2024Sage thoughts on the bizarrely parachuted in idea of an AGI Manhattan project in the US China Commission report from @deanwb.bsky.social www.hyperdimensional.co/p/questions-...hyperdimensional.coQuestions, Unasked and UnansweredThe US-China Commission Floats an AGI Manhattan Project 081
Dean W. Ball @deanwb.bsky.social · 19/11/2024The US-China Economic and Security Review Commission, a group/report commissioned by Congress, puts an explicit AGI Manhattan Project as their top US policy recommendation. 130
