
Currently thinking about AI alignment and welfare. I've also worked on theory and algorithms for Markov chains. applied math @ICMEStanford, physics @ucsc
San Francisco Bay Area
Joined January 2018
- Tweets79
- Following818
- Followers303
- Likes505
Alex Infanger retweeted
I resigned from Google DeepMind bc it broke its founding promise by selling AI to the military without restrictions against killer robots or mass spying.
For months, I worked to stop this but watched powerful ethicists and institutions choose silence.
Here's what happened. 🧵
Alex Infanger retweeted
Extremely cogent and well-written scenario, starting from the present, warning about Europe’s dangerous trajectory. I was in Paris for the AI Summit after DeepSeek r1, and I can attest that the level of delusion about what that model meant was exactly as this scenario describes.
Alex Infanger retweeted
I'm deeply concerned about Europe's future on AI. One of my biggest worries is our erosion of agency, our ability to stay relevant and fight for our values in a future where AI becomes a civilisationally important technology.
Myself, @DadaJudith , @bakkermichiel and others have written a scenario to outline a potential future we worry we are on track towards.
europe2031.ai/
Every optimistic and realistic path I can see for Europe runs through a central node - one where Europe has more leverage, more importance and more say. One where Europe grows more, builds more where it matters, and takes ownership over its resilience.
Europe 2031 is a five-year scenario of the continent's slide into irrelevance: how AI is driving it, and what can still be done.
The co-authors are researchers, scientists and investors who have advised European leaders, co-authored national AI strategies, built and funded these systems from the inside. We have no interest in hype and we deeply care about this continent.
Europe 2031 ends with five concrete recommendations:
- drastically more compute on European soil
- an AI middle-power coalition
- labour-market reforms
- a bold position in robotics and industrial AI
- and a positive vision of what AI can do for society.
Europe can still change course if it finds the political will and the courage to engage in the most ambitious political and economic agenda the continent has undertaken in peacetime.
I encourage you to read it if you have the time:
Alex Infanger retweeted
1/ Deep learning is going to have a scientific theory. We can see the pieces starting to come together, and it's looking a lot like physics!
We're releasing a paper pulling together these emerging threads and giving them a name: learning mechanics.
🔨 arxiv.org/pdf/2604.21691 🔧
I can't speak highly enough of my time at MATS with Team Shard. For those considering it, feel free to DM me to chat.
Come work with me and Alex Cloud this summer in Team Shard at MATS! We have fun, consistently make real alignment progress (we pioneered steering vectors in 2023!), and help scholars tap into their latent abilities.
4
281
Alex Infanger retweeted
Visit our poster at NeurIPS today!
- Exhibit Hall C, D, E #1103
- Wed, Dec 3, 2025 • 4:30 PM – 7:30 PM PST
Alex Infanger retweeted
So excited about the paradigm emerging for the past few years where models are dispensable artifacts—and where systems are compiled from much higher-level processes than just taking gradient steps.
Neat approach:
We did some mech interp theory work. Planning to build on this, so let us know what you think!
Check our new AI paper, The Persian Rug!
arxiv.org/abs/2410.12101
We extract exactly the algorithm learned by the most well known model of neural network superposition, and distilled it into a set of weights resembling a Persian rug, which matches the learned loss exactly.
3
452
Alex Infanger retweeted
Looking for something to check out on the last day of #NeurIPS2023? Come hang out with EleutherAI @solarneurips
@BlancheMinerva is speaking on a panel and @jacob_pfau @alexinfanger Abhay Sheshadri, Ayush Panda, Curtis Huebner and @_julianmichael_ have a poster
Room R06-R09
Alex Infanger retweeted
GPQA is now on Hugging Face! 🤗
huggingface.co/datasets/idav…
🧵Announcing GPQA, a graduate-level “Google-proof” Q&A benchmark designed for scalable oversight! w/ @_julianmichael_, @sleepinyourhat
GPQA is a dataset of *really hard* questions that PhDs with full access to Google can’t answer.
Paper: arxiv.org/abs/2311.12022
Alex Infanger retweeted
Thanksgiving?
more like Thanks for Giving me and your friends your wonderful presence and beautiful smile you silly goose
Alex Infanger retweeted
1) Vision: when I joined in 2020 I was impressed to hear a coherent plan (from @sama @ilyasut @gdb + others) to build AGI, one that's roughly panning out. leads to everyone pulling in the same direction. at other labs, it felt like people pull randomly
Alex Infanger retweeted
Jhourney's looking for some volunteers in the Bay Area interested in being our first guinea pigs next week for our earliest attempts at jhana-accelerating biofeedback.
Will require ~4 hours. We'll pull out everything we've got to try and teach you the jhanas.
For vibecampers still on the east coast and interested in consciousness. Note that this is happening in NYC this weekend:
theassc.org/assc-26/
1
285
Did anyone who went to @ewu_wtf's clowning workshop at vibecamp remember the name of the Brooklyn clown school they mentioned at the end? Was it "clown gym"?
1
2
227
Alex Infanger retweeted
Poincaré inequality gives bounds on a function using its derivative(s). For example, ‖f‖₂ ≤ c ‖∇f‖₂ for some constant c.
Turns out there's a Gaussian version too. e.g. If x ∼ 𝒩(0,C) is a Gaussian n-vector, and C is the covariance, then
Var[ f(x) ] ≤ 𝔼[ ∇fᵀ C ∇f ]
Alex Infanger retweeted
People worry about how much land we'd need to supply the US with clean energy. Well, @elonmusk and I have independently calculated it and we both come up with something roughly comparable to the area we currently use for maple syrup or golf. A square about 100-120 miles per side.
This map gives you a rough sense of all the ways U.S. land is used. Much of U.S. land serves specific purposes, such as the 2 million acres devoted to golf courses or the 3 million acres for airports.bloom.bg/37XnVYh
Alex Infanger retweeted
Planes made by Boeing and Airbus aren’t faster than they were in the 1970s, but they are 70% more efficient per passenger and mile. Even long-haul flights now beat cars for fuel efficiency per passenger/mile. The problem is more people are flying further.
wsj.com/articles/airplanes-c…
Alex Infanger retweeted
A mean field evolution is the large number limit of particles systems. The simplest case is gradient descent, which corresponds to an advection equation in the limit. en.wikipedia.org/wiki/Advect… en.wikipedia.org/wiki/Mean_f…

































