Sign in

Sebastian Benthall

@sbenthall.bsky.social
291 followers 577 following 291 posts

scientist, dad. working on neurosymbolic AI systems for computational economics. Research Scientist, International Computer Science Institute Research Director, New York University School of Law sbenthall.net - digifesto.com

PostsRepliesMedia
Sebastian Benthall @sbenthall.bsky.social · 9h
scikit-agent 0.2.0 is out. It adds multi-agent models: populations, aggregation, and equilibrium solvers, with new gallery examples of Cournot, Akerlof's lemons, differential privacy as a causal game, and AI-safety incentives. github.com/scikit-agent...
github.com
Release v0.2.0 · scikit-agent/scikit-agent
Multi-agent model representation Blocks can declare an entity class, such as a set of firms or a population of households, whose variables take one value per member. Equations that read across a c...
000
Sebastian Benthall @sbenthall.bsky.social · 10/09/2026
the potential of AI to cause human extinction (or similarly bad outcomes) AI X-risk is in the news cycle again. This time, it comes when AI companies have imminent IPOs and when their products are taking off. So while concerns about AI X-risk have been around for quite a while, and were part of…
digifesto.com
the potential of AI to cause human extinction (or similarly bad outcomes)
AI X-risk is in the news cycle again. This time, it comes when AI companies have imminent IPOs and when their products are taking off. So while concerns about AI X-risk have been around for quite a while, and were part of the motivating ideologies of many of the AI lab founders, we are being told (again) that this problem is especially urgent, now.
110
Sebastian Benthall @sbenthall.bsky.social · 27/08/2026
scikit-agent 0.1.0 released! Well over a year ago I wrote about building something new. I'm pleased to say that now there is a 0.1.0 release of scikit-agent available. Like one does, I find that as I work on it, what was a simple idea expands, and I now understand that the project is indeed quite…
digifesto.com
scikit-agent 0.1.0 released!
Well over a year ago I wrote about building something new. I'm pleased to say that now there is a 0.1.0 release of scikit-agent available. Like one does, I find that as I work on it, what was a simple idea expands, and I now understand that the project is indeed quite technically ambitious. This first release is just a tiny part of what the library should be.
100
Sebastian Benthall @sbenthall.bsky.social · 31/07/2026
brief theses on wealth and value There's a longer version of this argument to be made, but I wanted to see if I could present it briefly. What is wealth? Nominal financial worth -- in terms of, e.g. US dollars, and other fiat currency, is clearly not a good measure of wealth. Because it is often…
digifesto.com
brief theses on wealth and value
There's a longer version of this argument to be made, but I wanted to see if I could present it briefly. What is wealth? Nominal financial worth -- in terms of, e.g. US dollars, and other fiat currency, is clearly not a good measure of wealth. Because it is often based on speculative pricing. Because it is vulnerable to inflation.
110
Sebastian Benthall @sbenthall.bsky.social · 22/07/2026
The Open Silicon Fallacy: How Panic Over Chinese Open-Weight Models Is Distorting American AI Policy digifesto.com/2026/07/22/t...
digifesto.com
The Open Silicon Fallacy: How Panic Over Chinese Open-Weight Models Is Distorting American AI Policy
The following article was written with Gemini Flash 3.6 “in the style of The Economist’s Bagehot”. It is an experiment in AI writing. The arguments and structure are mine. –…
000
Sebastian Benthall @sbenthall.bsky.social · 04/07/2026
some philosophical progress and next steps digifesto.com/2026/07/04/s...
digifesto.com
some philosophical progress, and next steps
I had a wonderful time at FAccT 2026 in Montreal. I was at the first FAT* in 2018 and thought it was doing something important — creating an intellectual space for serious thinking about powe…
000
Sebastian Benthall @sbenthall.bsky.social · 02/01/2026
naturalized ethics and natural law digifesto.com/2026/01/02/n...
digifesto.com
naturalized ethics and natural law
One thing that’s become clear to me lately is that I now believe that ethics can be naturalized. I also believe that there is in fact a form of ‘natural law’. By this I mean that …
020
Sebastian Benthall @sbenthall.bsky.social · 19/12/2025
If I actually finish any of these 4-5 projects I'm working on, it's going to be so great.
000
Reposted by Sebastian Benthall
NYU Engelberg Center @nyuengelberg.org · 09/12/2025
Thank you KAIST for hosting our Legal Challenges in the AI Era workshop! @jeannefromer.bsky.social, @michaelweinberg.org, @sbenthall.bsky.social, & Florencia Marotta-Wurgler are representing NYU in Seoul
Florencia Marotta-Wurgler on a panelJeanne Fromer presentsSeb Benthall presentsJeanne Fromer picks winners from the lucky box
012
Reposted by Sebastian Benthall
Josh McCrain @joshmccrain.bsky.social · 18/11/2025
new paper by Sean Westwood: With current technology, it is impossible to tell whether survey respondents are real or bots. Among other things, makes it easy for bad actors to manipulate outcomes. No good news here for the future of online-based survey research
38770390
Sebastian Benthall @sbenthall.bsky.social · 15/11/2025
I'm very +1 the LeCun world model pivot.
000
Sebastian Benthall @sbenthall.bsky.social · 14/11/2025
Europe once again realizes that it can get a lot of money out of US big tech companies by enforcing laws
020
Reposted by Sebastian Benthall
Will Davies @will-davies.bsky.social · 11/11/2025
Great piece by Fred Turner on the pivot from Californian ideology of connection (which built the web) to Texan ideology of extraction and theocracy thebaffler.com/salvos/the-t...
thebaffler.com
1229
Reposted by Sebastian Benthall
Steve Randy Waldman @interfluidity.com · 10/11/2025
drafts.interfluidity.com/2025/10/04/a...
drafts.interfluidity.com
A fertile corpse
drafts @ interfluidity
1103
Sebastian Benthall @sbenthall.bsky.social · 11/11/2025
I was thinking today that we need something like Marr's levels, but for software system design. Level one is for functional specification. Level 2 technical design. Level 3 is implementation. Level 0 is incentives and politics of design. System errors can happen at all levels.
120
Sebastian Benthall @sbenthall.bsky.social · 10/11/2025
[2510.27628] Validity Is What You Need arxiv.org/abs/2510.27628
arxiv.org
Validity Is What You Need
While AI agents have long been discussed and studied in computer science, today's Agentic AI systems are something new. We consider other definitions of Agentic AI and propose a new realist definition...
010
Reposted by Sebastian Benthall
Eryk Salvaggio @eryk.bsky.social · 01/11/2025
RIP Alison Knowles. Damn. hyperallergic.com/1054017/alis...
hyperallergic.com
Alison Knowles, the First Woman of Fluxus, Dies at 92
“People don’t touch art,” she once said. “That’s one of the problems.”
0299
Sebastian Benthall @sbenthall.bsky.social · 29/10/2025
When will the Government shutdown end? share.google/2WvCZK7Lrzp0...
share.google
When will the Government shutdown end?
Polymarket | This market will resolve to the calendar date range (ET) of the first day after September 30, 2025 which the U.S. The Office of Personnel Manage...
010
Sebastian Benthall @sbenthall.bsky.social · 25/10/2025
Honestly relieved that my child fell asleep on the way to the Halloween parade and won't let himself be woken up so I too can nap in the car.
000
Sebastian Benthall @sbenthall.bsky.social · 20/10/2025
There are ever increasingly impressive demos of high-tech machines promising a transformed future. But somehow sustaining automation of even universal basics remains a challenge. Kitchen appliances break down. We haven't yet mastered the smoke alarm that doesn't go off when you fry stuff. Etc.
011
Sebastian Benthall @sbenthall.bsky.social · 19/10/2025
I'm glad the political symbolism of the frog changed. I was pretty sick of the ugly meme cartoon.
000
Sebastian Benthall @sbenthall.bsky.social · 13/10/2025
Today I saw _Robots_ (2005) for the first time. Free with ads on YouTube. Quite underrated children's movie. Brilliant animation. Mel Brooks and Robin Williams voice acting. My son enjoyed it.
000
Reposted by Sebastian Benthall
Catalin Cimpanu @campuscodi.risky.biz · 09/10/2025
Looks like California has learned something from the EU Cookie Banner disaster therecord.media/california-s...
text that reads:  California Gov. Gavin Newsom on Wednesday signed a bill which requires web browsers to make it easier for Californians to opt-out of allowing third parties to sell their data.

The California Consumer Privacy Act, signed in 2018, gave Californians the right to send opt-out signals, but major browsers have not had to make opt-outs simple to use. The bill signed Wednesday would require them to set up an easy-to-find mechanism that lets Californians opt-out with the push of a button, instead of having to do so repeatedly when visiting individual websites.
1228
Sebastian Benthall @sbenthall.bsky.social · 08/10/2025
AT Protocol hosts splitting away from Bluesky over content moderation politics is actually what the AT Protocol needs to achieve its objective of decentralization. Feature not bug.
000
Sebastian Benthall @sbenthall.bsky.social · 07/10/2025
Sociology is a martial art. youtu.be/PA4s3I3-X2U?...
youtu.be
Sociology is a Martial Art 2 new
YouTube video by Claes pålsson
010
Sebastian Benthall @sbenthall.bsky.social · 07/10/2025
Speaking of primatology, my main opinion about jobs and automation is that a large share of "economic activity" is elaborate social grooming to establish hierarchy, and that the share of human labor devoted to this will increase with automation, because it cannot be automated.
000
Reposted by Sebastian Benthall
Humanists UK @humanists.uk · 07/10/2025
Groundbreaking! The UK Armed Forces has sworn in its first-ever humanist pastoral carer – Dr Neil Weddell. This means thousands of non-religious service personnel will have access to like-minded pastoral support that genuinely reflects their values and worldview. 🧵
First-ever humanist pastoral carer in UK Armed Forces
410217
Sebastian Benthall @sbenthall.bsky.social · 01/10/2025
Baumol's cost disease, but one job. Everything that can be automated and streamlined about your work takes less and less time. The more efficient they get, the stagnant parts of one's work takes up relatively more time. Progress is blocked by the most tedious blockers. Is there a name for this?
110
Sebastian Benthall @sbenthall.bsky.social · 25/09/2025
Claim: true ethical statements refer to facts about which behaviors and institutions produce adaptively stable social systems, under varying environmental and technical regimes. These are sometimes unknown but largely discoverable facts.
010
Reposted by Sebastian Benthall
tante @tante.cc · 23/09/2025
"Low effort, unhelpful AI generated work is having a significant impact on collaboration at work. Approximately half of the people we surveyed viewed colleagues who sent workslop as less creative, capable, and reliable than they did before ...AI-Generated “Workslop” Is Destroying Productivity
1325149
Sebastian Benthall @sbenthall.bsky.social · 31/08/2025
So are CA, IL, and NY going to lead a shadow cabinet with all the fired federal experts developing state policy based on expertise? Then raise state taxes to match what was cut from federal?
010
Sebastian Benthall @sbenthall.bsky.social · 26/08/2025
Sovereign Wealth Funds can hold domestic equities. Ideological inconsistencies aside, the Intel thing seems like a fine or at least justifiable move. Cf. "Sovereign Wealth Funds as domestic investors of last resort during crises" | Cairn.info share.google/3JfOQmVu97BQ...
cairn.info
Cairn.info
Plateforme de référence pour les publications de sciences humaines et sociales
010
Sebastian Benthall @sbenthall.bsky.social · 22/08/2025
Does anybody have an leads or ideas about policies for open source software communities about the use of GenAI in contributions or participation?
000
Sebastian Benthall @sbenthall.bsky.social · 18/08/2025
Sure wish people would listen to the experts saying how important it is to trust the institutions with the experts in them. Timely PSA.
061
Sebastian Benthall @sbenthall.bsky.social · 17/08/2025
In neoclassical economic modeling, the assumption of rational decision-making, and the aggregation of multiple actors into a representative agent, are not orthogonal. By taking both steps, the model sheds its ability to represent the sections of society that are less rational and thereby diminished.
000
Sebastian Benthall @sbenthall.bsky.social · 16/08/2025
There are lots of reasons not to like Vladimir Putin. But I think I would actually prefer him as a President to who we currently have. Fewer embarrassing tweets, etc.
010
Reposted by Sebastian Benthall
Hubert Sablonnière @hsablonniere.com · 13/08/2025
🤓 Fascinating read about OpenAI, Anthropic and the business of LLMs ethanding.substack.com/p/openai-bur...
ethanding.substack.com
openai burns the boats
the $334 machine that openai is aiming at anthropic
2259
Sebastian Benthall @sbenthall.bsky.social · 12/08/2025
"Oligarchy" is a nice, fitting word because it implies both a concentration and multipolarity of power, unencumbered by totalizing ideology or even claim to personal virtue.
010
Reposted by Sebastian Benthall
Sebastian Benthall @sbenthall.bsky.social · 12/08/2025
A difference between the Trump epistemology and the Party epistemology in 1984 is that in the latter, there was unified and singular totalitarian control over records, ideology, belief, perception. There's a collectivist solipsism, ergo "Reality" is at stake. We have absolutely nothing like that.
131
Reposted by Sebastian Benthall
lastpositivist.bsky.social @lastpositivist.bsky.social · 09/08/2025
Quite the broadside from Stoller against the Abundance crowd, and I had a bit of a jumpscare to see effective altruism return to the Discourse too. There's an underlying point about to be made, like, what sort of things fool what sort of people. www.thebignewsletter.com/p/an-abundan...
thebignewsletter.com
An Abundance of Sleaze: How a Beltway Brain Trust Sells Oligarchy to Liberals
How did a debate over housing become a call to end the anti-monopoly movement? Abundance co-author Derek Thompson used slimy tactics to protect wealth and power in America. It's worth examining how.
25011
Sebastian Benthall @sbenthall.bsky.social · 10/08/2025
The Workshop on Law-Following AI this past week in Cambridge was excellent. Very interesting kernel of a community around AI compliance issues.
110
Sebastian Benthall @sbenthall.bsky.social · 04/08/2025
Traveling in the UK. It's comparatively quite nice over here.
000
Reposted by Sebastian Benthall
aaron from queens from twitter @aaronnarraph.bsky.social · 02/08/2025
Zohran didn't win by relying on standard turnout patterns. We had to fundamentally transform the Democratic electorate. I was part of a group of NYC-DSA members that applied lessons from our previous campaigns to design the “universe” of voters to talk to. Here's how we won 🧵
341634432
Reposted by Sebastian Benthall
Tobias Gerstenberg @tobigerstenberg.bsky.social · 02/08/2025
Josh Tenenbaum's inspiring keynote at #cogsci2025 on growing vs scaling AI, the big questions of cognitive science, and the many open questions for the field.
16414
Sebastian Benthall @sbenthall.bsky.social · 28/07/2025
I have been thinking about "National Brotherhood Week" for a couple of months now, so today's news hits hard.
110
Reposted by Sebastian Benthall
Al Yankovic @alyankovic.bsky.social · 27/07/2025
My last living musical hero is still my hero but unfortunately no longer living. RIP to the great, great Mr. Tom Lehrer.
Photo of Tom Lehrer
816280004710
Sebastian Benthall @sbenthall.bsky.social · 22/07/2025
The 99-1 flop of the restriction on state AI laws in the BBB shows that the industry is not as (legislatively) powerful as they seemed at the start of this year.
020
Reposted by Sebastian Benthall
Thomas Dietterich @tdietterich.bsky.social · 20/07/2025
Beautifully succinct statement of the core problem with the blind application of statistical learning in AI!!
A model trained with a frequentist objective using i.i.d. observational train and validation data is almost surely converging to the non-causal model of the underlying data generating process.

That is because using short cuts/spurious correlations leads to a lower loss than not, i.e. you are actually specifically incentivizing your model not to be causal.

"Emergence" believers' core fallacy is the assumption the causal model would have the lowest error, and therefore better optimization would eventually converge to the causal model. 

The opposite is true - the best scoring model (per frequentist objective) is actually the one that is best at exploiting spurious correlations.
5487
Sebastian Benthall @sbenthall.bsky.social · 18/07/2025
Ok, GenAI is useful for coding. CoPilot's pull request reviews on GitHub are surprisingly good. Asking specific questions about solutions to problems, where some conceptual idea needs an explanation, is useful. There's a lot of ways to use it that aren't "vibe coding" at all.
010