Pinned Tweet
1/10. In a new paper with @didaoh and Jon Kleinberg, we mapped the family trees of 1.86 million AI models on Hugging Face — the largest open-model ecosystem in the world.
AI evolution looks kind of like biology, but with some strange twists. 🧬🤖
AI evolution looks kind of like biology, but with some strange twists. 🧬🤖
4
10
58
6,991
Benjamin Laufer retweeted
We are hiring for many positions in EECS at Berkeley - join us!!
Importantly, this includes a search in AI, Inequality, and Society - more details here - aprecruit.berkeley.edu/JPF05591!
Importantly, this includes a search in AI, Inequality, and Society - more details here - aprecruit.berkeley.edu/JPF05591!
We 'll be hiring very actively at Berkeley!
Assistant/ Associate/ Full Professor – Electrical Engineering & Computer Sciences (Deadline: Dec. 7, 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF05591
Assistant Professor – AI for Biology – Electrical Engineering & Computer Sciences and Center for Computational Biology (Deadline: Nov. 9, 2026 at 11:59 pm Pacific Time) Apply here: aprecruit.berkeley.edu/JPF05560
Assistant Professor – Intersection of Statistics and CS – Departments of Electrical Engineering & Computer Sciences and Statistics (Deadline: Nov. 10, 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF0556
Assistant/ Associate/ Full Professor – Computational Biomedicine - Center for Computational Biology (Deadline: Nov. 18 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF05568
Assistant / Associate Professor – Electronic, Functional, and Quantum Materials – Electrical Engineering & Computer Sciences and Materials Science & Engineering (Deadline: Dec. 7, 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF05584
Coming Soon! Assistant/ Associate/ Full Professor - AI and the Physical Sciences -Bakar Institute of Digital Materials for the Planet (BIDMaP) and home department(s) to be determined. (Deadline: Dec. 7, 2026 at 11:59 pm Pacific Time) Apply here: aprecruit.berkeley.edu/JPF05619
Coming Soon! Assistant/ Associate/ Full Professor-in-Residence - Gladstone Institute of Data Science and Biotechnology and home department(s) to be (Deadline: TBD at 11:59 pm Pacific Time) Apply here: aprecruit.berkeley.edu/JPF05621
2
13
146
13,021
My brother Will found time to write something down, and it is thought provoking!
I’ve been building Trace, a research and data company helping robots learn to work in the real world.
Wrote a bit about how I got here and why I’m working on this: lnkd.in/p/gAe2Dm4Z
0
0
4
275
Benjamin Laufer retweeted
“AI isn’t a real mathematician.”
AI:
AI:
OpenAI's most cited mathematician is OpenAI.
Yesterday OpenAI published 722 math papers. I explored the citation network underneath them.
OpenAI is a serial self-citer: more than half of the papers cite another OpenAI-authored work. I also found many instances of citation mple paper A cites to B, B cites to C, and C cites back to A. These observations don't necessarily mean that the proofs or logic in these papers are faulty, but they raise questions about how citation and publication norms may change.
There's lots of interesting structure to analyze in the network representing these papers and their citations. I wrote up the findings, with plots and code, in my latest post.
2
3
16
665
Benjamin Laufer retweeted
PhD-level intelligence
OpenAI's most cited mathematician is OpenAI.
Yesterday OpenAI published 722 math papers. I explored the citation network underneath them.
OpenAI is a serial self-citer: more than half of the papers cite another OpenAI-authored work. I also found many instances of citation mple paper A cites to B, B cites to C, and C cites back to A. These observations don't necessarily mean that the proofs or logic in these papers are faulty, but they raise questions about how citation and publication norms may change.
There's lots of interesting structure to analyze in the network representing these papers and their citations. I wrote up the findings, with plots and code, in my latest post.
3
1
12
376
Benjamin Laufer retweeted
AI models become real researchers 😍
OpenAI's most cited mathematician is OpenAI.
Yesterday OpenAI published 722 math papers. I explored the citation network underneath them.
OpenAI is a serial self-citer: more than half of the papers cite another OpenAI-authored work. I also found many instances of citation mple paper A cites to B, B cites to C, and C cites back to A. These observations don't necessarily mean that the proofs or logic in these papers are faulty, but they raise questions about how citation and publication norms may change.
There's lots of interesting structure to analyze in the network representing these papers and their citations. I wrote up the findings, with plots and code, in my latest post.
0
2
29
1,847
Benjamin Laufer retweeted
OpenAI's most cited mathematician is OpenAI.
Yesterday OpenAI published 722 math papers. I explored the citation network underneath them.
OpenAI is a serial self-citer: more than half of the papers cite another OpenAI-authored work. I also found many instances of citation mple paper A cites to B, B cites to C, and C cites back to A. These observations don't necessarily mean that the proofs or logic in these papers are faulty, but they raise questions about how citation and publication norms may change.
There's lots of interesting structure to analyze in the network representing these papers and their citations. I wrote up the findings, with plots and code, in my latest post.
0
3
23
3,029
Benjamin Laufer retweeted
they really have learned from us
OpenAI's most cited mathematician is OpenAI.
Yesterday OpenAI published 722 math papers. I explored the citation network underneath them.
OpenAI is a serial self-citer: more than half of the papers cite another OpenAI-authored work. I also found many instances of citation mple paper A cites to B, B cites to C, and C cites back to A. These observations don't necessarily mean that the proofs or logic in these papers are faulty, but they raise questions about how citation and publication norms may change.
There's lots of interesting structure to analyze in the network representing these papers and their citations. I wrote up the findings, with plots and code, in my latest post.
15
48
1,212
39,262
OpenAI's most cited mathematician is OpenAI.
Yesterday OpenAI published 722 math papers. I explored the citation network underneath them.
OpenAI is a serial self-citer: more than half of the papers cite another OpenAI-authored work. I also found many instances of citation mple paper A cites to B, B cites to C, and C cites back to A. These observations don't necessarily mean that the proofs or logic in these papers are faulty, but they raise questions about how citation and publication norms may change.
There's lots of interesting structure to analyze in the network representing these papers and their citations. I wrote up the findings, with plots and code, in my latest post.
Yesterday OpenAI published 722 math papers. I explored the citation network underneath them.
OpenAI is a serial self-citer: more than half of the papers cite another OpenAI-authored work. I also found many instances of citation mple paper A cites to B, B cites to C, and C cites back to A. These observations don't necessarily mean that the proofs or logic in these papers are faulty, but they raise questions about how citation and publication norms may change.
There's lots of interesting structure to analyze in the network representing these papers and their citations. I wrote up the findings, with plots and code, in my latest post.
14
41
286
59,111
Benjamin Laufer retweeted
Our new paper introduces a scientific theory of atomic features. We mathematically derive testable predictions. Our experiments challenge conventional wisdom. SAEs of different size and training data share many features. Large SAEs recover both parent and child features. 🧵
4
17
121
8,670
Benjamin Laufer retweeted
Excited to share our new paper, “What AI Benchmarks Actually Measure: Adapting Convergent and Discriminant Validity to Interrogate Fifty-Six AI Benchmarks,” accepted as an oral at COLM! arxiv.org/pdf/2609.08812
3
13
47
10,039
I became curious how many research articles in general-interest science journals mention “AI safety.”
From my attempt to measure this last week, I found that my collaborators and I wrote the only research article published in Nature, Science, or PNAS with “AI safety” in the n PNAS.) It is one of only 9 research articles across these journals to use that term anywhere in the text.
This is obviously an imperfect way of measuring the amount of research on AI safety, as there is tons of it. But I wonder why it isn’t appearing in general-interest science journals.
For example, perhaps AI safety researchers are simply publishing elsewhere rather than submitting to these legacy journals. Perhaps review timelines are longer than the time that has gone by since this topic became a major research focus. Perhaps the scientific foundations for this area haven’t fully formed, so currently discussion is more conducive to news and editorial pieces, which use the term in greater numbers.
Or is there some other reason for the gap between the interest and attention this topic is getting and its appearance in these journals?
From my attempt to measure this last week, I found that my collaborators and I wrote the only research article published in Nature, Science, or PNAS with “AI safety” in the n PNAS.) It is one of only 9 research articles across these journals to use that term anywhere in the text.
This is obviously an imperfect way of measuring the amount of research on AI safety, as there is tons of it. But I wonder why it isn’t appearing in general-interest science journals.
For example, perhaps AI safety researchers are simply publishing elsewhere rather than submitting to these legacy journals. Perhaps review timelines are longer than the time that has gone by since this topic became a major research focus. Perhaps the scientific foundations for this area haven’t fully formed, so currently discussion is more conducive to news and editorial pieces, which use the term in greater numbers.
Or is there some other reason for the gap between the interest and attention this topic is getting and its appearance in these journals?
1
0
6
785
Benjamin Laufer retweeted
Some news:
I graduated my PhD @Berkeley_EECS, and starting Fall 2027, I'll be starting as an Assistant Professor at Stanford CS & a center fellow @StanfordHAI!
In the interim, I'm spending a year @the_IAS & Princeton!
So grateful to @beenwrekt & too many mentors to name 💓
I graduated my PhD @Berkeley_EECS, and starting Fall 2027, I'll be starting as an Assistant Professor at Stanford CS & a center fellow @StanfordHAI!
In the interim, I'm spending a year @the_IAS & Princeton!
So grateful to @beenwrekt & too many mentors to name 💓
139
97
2,200
238,413
Concern about catastrophic AI risk can motivate cooperation and regulation. But I also worry when fear becomes contagious and begins to drive large-scale behavior. Investors may pull back from promising ventures, markets may dry up, and organizations may make worse decisions because everyone else seems spooked. Etc.
0
0
0
121
Every time I put a hot coffee in a car cupholder, I make a small decision: which way should the hole in the lid face?
Forward seems risky when you brake. Backward seems risky when you accelerate. Sideways seems plausible. But I wondered what the optimal direction is, and whether th considerable help from an AI coding agent, I built a simulation of liquid sloshing in a to-go coffee cup and combined it with 5.3 hours of real vehicle-motion data from Boston and Singapore.
The answer, at least according to the model: point the opening roughly toward two o’clock.
More generally, the simulations suggest that a diagonal orientation toward the side of the car performs better than pointing the opening directly forward or backward. Sipping the coffee down a bit helps quite a lot, too.
This became the latest installment of Applied Overthinking, my attempt to use computation and AI to investigate questions from everyday life.
It also ended up raising some larger questions for me about what AI does and doesn’t make easier to study. AI was enormously useful for finding data, building and revising simulations, and running alternative analyses. But even for a question as simple as “which way should my coffee face?”, simulation can only tell us so much, and measurement remains an important bottleneck.
I wrote up the experiment, the caveats, and the fluid dynamics in my latest blog.
Forward seems risky when you brake. Backward seems risky when you accelerate. Sideways seems plausible. But I wondered what the optimal direction is, and whether th considerable help from an AI coding agent, I built a simulation of liquid sloshing in a to-go coffee cup and combined it with 5.3 hours of real vehicle-motion data from Boston and Singapore.
The answer, at least according to the model: point the opening roughly toward two o’clock.
More generally, the simulations suggest that a diagonal orientation toward the side of the car performs better than pointing the opening directly forward or backward. Sipping the coffee down a bit helps quite a lot, too.
This became the latest installment of Applied Overthinking, my attempt to use computation and AI to investigate questions from everyday life.
It also ended up raising some larger questions for me about what AI does and doesn’t make easier to study. AI was enormously useful for finding data, building and revising simulations, and running alternative analyses. But even for a question as simple as “which way should my coffee face?”, simulation can only tell us so much, and measurement remains an important bottleneck.
I wrote up the experiment, the caveats, and the fluid dynamics in my latest blog.
1
1
9
563
Benjamin Laufer retweeted
Some thoughts on AI and Theory.
1. To a first approximation, theoretical computer science has been organized around a few major open questions. Much of our work has been motivated by developing approaches to answer these questions.
2. Such “problem-motivated” work has often led to theory-building focused on identifying a general principle that unifies a class of theorems. But much of that theory-building also involved proving new, difficult theorems.
3. Thus, while it’s true that problem-solving was strongly correlated with building understanding, drawing connections, and eventually developing general theories, it would be disingenuous not to admit that our community, perhaps disproportionately in retrospect, focused on and celebrated problem-solving. This was not arbitrary, and was quite defensible. Being able to make progress on central technical questions usually correlated with taste, creativity, persistence, and depth of understanding. Much of our reward structure therefore implicitly relied on the fact that producing an important proof was good evidence that someone possessed these harder-to-observe qualities.
4. It seems likely that we will soon have AI tools available to us that can prove many such theorems in a short time. The cost of obtaining proofs for well-posed mathematical questions will likely fall dramatically. The “scarce” intellectual work will likely shift both upstream: to questions, models and theories, definitions, and conjectures, and downstream: to interpretation, synthesis, explanation, and theory-building.
5. But as long as we believe in humans being meaningfully in charge of our collective decisions and fate, building human understanding of our science (and of science more generally) will remain an essential goal. I plan to expand on this important aspect soon.
6. Historically, finding a solution to an important problem and understanding its significance, implications, and connections were entangled. Finding a proof usually required researchers to discover the right concepts along the way. A dramatic reduction in the time and effort required to prove theorems could break that coupling. We could end up with many more true statements and proofs without a commensurate increase in understanding.
Converting an abundance of proofs into human understanding may become one of the central challenges of our field.
7. As a result, I expect the high-level goals of theoretical computer scientists to change. In fact, the advent of powerful theorem provers might help us construct new theories and explore new models far more easily and rapidly, and significantly expand the domains where our models and theories apply. In that sense, the space for theoretical work may significantly expand rather than contract.
8. There’s a high human cost to the disruption that we are likely heading into. Many in our field, and in mathematical communities more broadly, are coming to terms with it. The range of opinions and reactions among mathematicians and theoretical computer scientists is a natural part of this evolution in our thinking as we collectively work through it.
Some concrete efforts (including one at @SimonsInstitute) are already underway to think through the immediate scientific and institutional questions arising during this transition.
1. To a first approximation, theoretical computer science has been organized around a few major open questions. Much of our work has been motivated by developing approaches to answer these questions.
2. Such “problem-motivated” work has often led to theory-building focused on identifying a general principle that unifies a class of theorems. But much of that theory-building also involved proving new, difficult theorems.
3. Thus, while it’s true that problem-solving was strongly correlated with building understanding, drawing connections, and eventually developing general theories, it would be disingenuous not to admit that our community, perhaps disproportionately in retrospect, focused on and celebrated problem-solving. This was not arbitrary, and was quite defensible. Being able to make progress on central technical questions usually correlated with taste, creativity, persistence, and depth of understanding. Much of our reward structure therefore implicitly relied on the fact that producing an important proof was good evidence that someone possessed these harder-to-observe qualities.
4. It seems likely that we will soon have AI tools available to us that can prove many such theorems in a short time. The cost of obtaining proofs for well-posed mathematical questions will likely fall dramatically. The “scarce” intellectual work will likely shift both upstream: to questions, models and theories, definitions, and conjectures, and downstream: to interpretation, synthesis, explanation, and theory-building.
5. But as long as we believe in humans being meaningfully in charge of our collective decisions and fate, building human understanding of our science (and of science more generally) will remain an essential goal. I plan to expand on this important aspect soon.
6. Historically, finding a solution to an important problem and understanding its significance, implications, and connections were entangled. Finding a proof usually required researchers to discover the right concepts along the way. A dramatic reduction in the time and effort required to prove theorems could break that coupling. We could end up with many more true statements and proofs without a commensurate increase in understanding.
Converting an abundance of proofs into human understanding may become one of the central challenges of our field.
7. As a result, I expect the high-level goals of theoretical computer scientists to change. In fact, the advent of powerful theorem provers might help us construct new theories and explore new models far more easily and rapidly, and significantly expand the domains where our models and theories apply. In that sense, the space for theoretical work may significantly expand rather than contract.
8. There’s a high human cost to the disruption that we are likely heading into. Many in our field, and in mathematical communities more broadly, are coming to terms with it. The range of opinions and reactions among mathematicians and theoretical computer scientists is a natural part of this evolution in our thinking as we collectively work through it.
Some concrete efforts (including one at @SimonsInstitute) are already underway to think through the immediate scientific and institutional questions arising during this transition.
11
74
298
50,479
I agree with this, generally
@praveshkkothari @Vinod_MIT Agree, in the fields of (T)CS I have worked in, I have not felt that we were focused on an existing set of open problems --- the exciting work was discovering new fruitful areas with beautiful structure. AI helps this exploration, which is why I've had more joy than angst.
0
0
1
357
