dan bateyko (@dbateyko) header
dan bateyko (@dbateyko) profile picture

maybe the hard stuff's inside, hidden — like bones, as opposed to an exoskeleton. @CornellInfoSci

NY
Joined January 2012
Photo by dan bateyko (@dbateyko)Photo by dan bateyko (@dbateyko)Photo by dan bateyko (@dbateyko)Photo by dan bateyko (@dbateyko)Photo by dan bateyko (@dbateyko)Photo by dan bateyko (@dbateyko)Photo by dan bateyko (@dbateyko)Photo by dan bateyko (@dbateyko)Photo by dan bateyko (@dbateyko)Photo by dan bateyko (@dbateyko)
anna’s archive pdf: formatted like a real book, easily accessible, citations hyperlinked actual academic press ebook reader: no the login through institution button doesn’t log you in. and if you do get in you can only navigate through the book reading interface in gta
21
1,109
19
11,732
131,917
dan bateyko retweeted
I love connecting people. I love starting little ___ <> ___ Signal chats. I think this is 80% of good policy work.
13
6
4
196
8,209
dan bateyko retweeted
please please please come work with me: openai.com/careers/research-… our team of (3) helps define, drive and operationalize so many of the hot button topics - embedded evaluators? check! misalignment incidents and reporting? check! 3p/government testing and frameworks? check! how these things interact with policy initiatives? check! @_lamaahmad is a hero and me and serena are at least pleasant to work with, and our stable of dedicated daily collaborators from @fouadmatin @EstherTetruas @j_nitishinskaya @kaicathyc @Eric_Wallace_ @dkaushik96 @MicahCarroll @Marcus_J_W @davideoks @kevin_klyman is also an all-timer crew
31
50
10
492
52,382
I'm excited to finally announce my new Substack! I'll be writing on AI evaluations & incidents, technical explainers, and more. Check out post #1: A primer for policy folks on what AI evaluations are, and why policymakers should care about them ai.scal.ing/p/evals-intro-1
8
21
108
2,835
dan bateyko retweeted
I am recruiting students at UW! I’m interested in AI ecosystems, evaluation and safety. If you are interested in doing a PhD with me, there are two routes: 1. Apply to UW iSchool PhD (due December 2!) and list my name in your application/SOP. 2. Apply to UW CSE PhD (due December 15) and list both Jamie Morgenstern and my names in your application/SOP. We’re recruiting students to co-advise. You can also apply to both of these, of course!
15
58
4
309
18,935
I’m also hiring for my team! Join us on Frontier Assurance to help make these assessments go well. openai.com/careers/research-…
1
4
36
3,052
The current Jensen Huang reaction reminds me of the Jacob Coxon reaction a few months ago -- Everyone paying attention pre-Coxon already knew that AI companies don't have a good plan for making superintelligence safe but are rushing ahead anyway, and that there's competitive pressure to keep going. Those people were glad that suddenly more people cared. Likewise, everyone paying attention knew that NVIDIA captured Commerce long ago and that tons of GPU smuggling, loopholes, etc. were being ignored as a result. Again, now experts are thinking "finally something might be done"! Something dramatic caught the public's attention, and the long-known issue became salient. Here are two other AI-related things that everyone paying attention knows but most people don't, which may or may not eventually become salient to the public: - There are no minimum safety and security standards in AI whatsoever. The law of the land is "publish your own policy and follow it." As one might expect, company policies are generally weak/non-committal as a result. Things like "don't train your model in a way that will predictably cause it to become hard to monitor (other than as part of safety research)" is not required, nor is monitoring AI agents' chains of thought for deceptive or malicious behavior, or taking serious measures to protect your IP from theft, etc. - Since no federal AI whistleblower protection law has passed, there is legal ambiguity around whistleblowing in AI safety and security. If a company employee shares confidential information with a member of Congress about something that they consider dangerous, they may not be protected from retaliation if the information they share doesn't involve violating any law. Of course some people still share such things, but they are taking risks doing so, and Congress is systematically under-informed as a result, just as it ostensibly is gearing up to regulate this technology.
12
41
6
301
16,704
At @AVERIorg, we’re hiring an Audit Manager. AVERI’s mission is to ensure that the most powerful AI systems, and the companies that build them, are rigorously audited for safety and security. To do so, we run pilot project which break new ground, and help us develop tools and processes to facilitate the nascent discipline of frontier AI auditing. We have run pilot projects with numerous frontier AI developers, most recently Google DeepMind (the world’s first double blind evaluation) and Amazon (scrutinizing the quality of their internal benchmarks). We are particularly interested in candidates who have either experience in audit and assurance, or frontier AI evaluations, as well as a track record of project or programme management. We offer a competitive salary (150-260k USD), comprehensive benefits including a yearly work enablement stipend, and most importantly a small, mission driven team committed to a critical mission - ensuring that frontier AI systems are developed safely and securely. Apply here: averi.org/careers?ashby_jid=…
2
3
1
18
1,017
dan bateyko retweeted
I'm giving a talk and answering questions on Tort Law for Frontier AI tomorrow at @Harvard_Law. Please come say hi! hls.harvard.edu/events/tort-…
2
9
2
67
2,859
Apply to be a postdoc at Cornell (NYC and Ithaca)! I am recruiting candidates interested in technical approaches to responsible AI.
Cornell (and me) is hiring postdocs, with focus on technical AI/ML research. Deadline for full consideration is Dec 15. I am recruiting through this call as well. I encourage applying as early as possible, because I am starting to review folders now! Application link ⬇️
6
45
2,718
dan bateyko retweeted
We released LegalBench in August 2023. In the three years since, it's become one of the more widely used benchmarks for legal AI. In the spirit of critical self-reflection, I wrote some thoughts about (1) what went well, (2) what I wish I'd done differently, and (3) what this all means for the future of legal AI research. I wanted to highlight a few specific reflections.
9
27
8
153
18,931
dan bateyko retweeted
Dear dozens of new economic populist members of Congress, You’ll face fast pressure to deliver economic results. How to do it? Subpoena CEOs. A shocking amount of time, they'll change what they are doing long before you pass a bill or actually cut off their federal funding. So: - Subpoena Google for cramming AI down our kids throats. Threaten to amend the FAR to prohibit contracting across the U.S. government with any company forcing kids to use AI. - Subpoena the private equity managers enshittifying hospitals, dialysis centers, your local vet, your local primary care provider. Name individuals, not just companies. Shame them. Show the public (and public pension funds) how they make their money. - Subpoena health insurers who use AI to auto-deny valid claims, triage mental health emergencies, or upcode diagnoses. Make them stop that. - Subpoena the airlines for systematically denying cash refunds to travelers. (“Here’s a voucher instead!”) Show your hand – threaten the subpoena now, before Thanksgiving. - Subpoena Brazilian meatpacker JBS for literally anything. For child labor, for getting fat and rich off of cattlemen, for making their workers work so fast their hands curl into claws, for persistent hygiene allegations. - Subpoena corporate landlords. Show the public how they get rich off of hidden fees and not fixing their apartments. - Subpoena any company that has been slow to issue tariff refunds to small business owners. I bet they’ll get their checks a lot quicker! Will it work? See what happened when Instacart was exposed as secretly marking up your grocery cart... Yes, it can work. P.S. If anyone tells you “No, you can’t do that” or “No, there’s a process for that,” ignore them. You know what time it is.
50
955
70
3,023
83,499
dan bateyko retweeted
wrote a Media History of the Infinite Feed for @kernel_magazine #6 back in august, out today in the @reboot_hq substack! it's a story about protocols, dynamic updates, AJAX and the democratic dream of social media. thanks to @Jacobkupp for edits! 🖨️📜 substack.com/home/post/p-222…
9
76
2,583
longer thing on this incoming, but i think there's a strong argument for the third wave of american philanthropy funding third spaces
friends with upcoming IPOs looking to fund local public goods in San Francisco - humbly requesting more beautiful, spacious, cozy cafes. coffee bar + library, lots of plants, warm lighting, turns into a wine bar in the evening
2
2
20
1,199
Today, most insurer exposure to AI agent risk sits unpriced within existing cyber, professional liability, and general liability policies. This summer, @gabriel_weil, Non-Resident Senior Fellow at LawAI and Assistant Professor at the University of Houston Law Center, alongside colleagues from Anthropic, OpenAI, RAND, Stanford, Oxford, and others, co-authored a report arguing that affirmative AI coverage with limits in the billions may be achievable by 2030. “Underwriting the Agent Economy: The Blueprint for an AI Insurance Stack” maps an eight component stack, including incident reporting, contract design, and pricing, and encourages the insurance industry to coordinate in order to manage AI risk adequately. Link to the report in the comments.
3
6
15
706
dan bateyko retweeted
I wrote about my experience on the industry job market as a PhD student in an AI-adjacent field (HCI): kjfeng.me/blog/industry-job-…. If you’re wrapping up your PhD, didn’t focus on LLM training, and thinking about industry jobs, this one’s for you!
9
54
2
408
36,620
Alex crushed it. The paper introduces new ways to measure how hard ideas are to articulate to AI. Shared social norms are especially difficult. Lots to explore for law-following AI e.g. how do we convey why an ambulance should be exempt from a park's ban on vehicles?
So excited for this research! I've always been annoyed at the supremacy of verified rewards, and now we can finally show it. In this work, we develop ways of measuring: how much of a reward signal is verifiable, how much can we put into words, and how much is shared but tacit, or impossible to articulate?
145
Extensions, ongoing and future work: Anyone who talked to me during this period probably heard a lot about “articulabilty gaps” but the more I started looking at them, the more I started seeing them everywhere. I'll talk about some of these future directions now. I’m convinced that tacit knowledge is everywhere, and is the key missing variable for making progress on frontier tasks. I have a number of follow-up projects, please reach out!
1
3
1
17
873
🚨New work: A SOTA prompt optimization framework + a novel unsupervised measure of prompt performance. Written rubrics and constitutions are crucially important for AI task improvement, safety following, pluralism and adaptation. But some of the things we might want a rubric to do are difficult to put into words; how would you even describe “comic voice”? I’m human, albeit not that funny, and I'm totally not sure what that means. What kinds of written constructs can you expect your judge to do well on and how good can you expect it to be?
6
20
2
91
10,942
dan bateyko retweeted
Yes, administrative lawyers, we need your insights for AI governance. New via @YaleJREG
2
4
1
23
1,013