Sign in

Simons Institute for the Theory of Computing

@simonsinstitute.bsky.social
1.8K followers 820 following 360 posts

The world's leading venue for collaborative research in theoretical computer science. Follow us at YouTube.com/SimonsInstitute.

PostsRepliesMedia
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 30/09/2026
Earlier this month, we hosted a working group to think about how the TCS community should respond to AI progress. Their report, released on Monday, highlights 12 near-term actions that received broad support among the participants. simons.berkeley.edu/news-publica...
051
Reposted by Simons Institute for the Theory of Computing
Clément Canonne @ccanonne.github.io · 29/09/2026
"Well, that's just, like, your opinion, man." (remotely attending the Simons Institute's Karp Distinguished Lecture by @booleananalysis.bsky.social) @simonsinstitute.bsky.social
Ryan O'Donnell's (almost) favorite expander graph
2213
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 30/09/2026
Earlier this month, we hosted a working group to think about how the TCS community should respond to AI progress. Their report, released yesterday, highlights 12 near-term actions that received broad support among the participants. simons.berkeley.edu/news-publica...
001
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 25/09/2026
4/4 "This allows you to interpolate between auto-regression and diffusion," said Volodymyr Kuleshov of @cornelluniversity.bsky.social at the Simons Institute workshop on Diffusion Generative Modeling: Progress and Next Steps. Video: simons.berkeley.edu/talks/volody...
simons.berkeley.edu
How to Build a Modern Diffusion Language Model
This talk introduces diffusion for language and the research advances that underlie modern diffusion LLMs. We describe the building blocks of today's leading open-source models, starting from simple m...
020
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 25/09/2026
3/4 Block diffusion addresses this limitation. "You'd run diffusion over blocks of tokens in parallel and generate the following block of tokens, using a fixed length diffusion block, conditioned on the [generated] tokens," said V. Kuleshov of @cornelluniversity.bsky.social at the Simons Institute
110
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 25/09/2026
2/4 For e.g., current diffusion language models have "been defined to only generate fixed length sequences," whereas autoregressive models can keep adding tokens and generate sequences of variable length, said Volodymyr Kuleshov of @cornelluniversity.bsky.social at the Simons Institute
110
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 25/09/2026
1/4 Are diffusion language models ready for the real world? Not quite. Today's diffusion language models are missing some key ingredients, said Volodymyr Kuleshov of @cornelluniversity.bsky.social, at the Simons Institute workshop on Diffusion Generative Modeling: Progress and Next Steps
120
Reposted by Simons Institute for the Theory of Computing
Clément Canonne @ccanonne.github.io · 22/09/2026
Quite the exciting talk by Ryan O'Donnell (@booleananalysis.bsky.social) at the Simons Institute next week, on Expander Graphs. We'll finally know which ones are his favorites! Livestream option available upon (free) registration: simons.berkeley.edu/events/my-fa... @simonsinstitute.bsky.social
simons.berkeley.edu
My Favorite Expanders | Richard M. Karp Distinguished Lecture
Expander graphs are sparse graphs for which, whenever you split the graph into two parts, the number of edges going between the parts is proportional to the size of the smaller part. There are extreme...
1132
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 23/09/2026
Join us Tuesday, 9/29 for the first Richard M. Karp Distinguished Lecture of this academic year, featuring @booleananalysis.bsky.social. simons.berkeley.edu/events/my-fa...
063
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 23/09/2026
We invite your project proposals for Circles, the Simons Institute – Jane Street Small Group Collaborations, which supports groups of 3-6 researchers for 4 weeklong gatherings over 2 years. Apply by Oct. 15 (deadline extended). simons.berkeley.edu/participate/...
033
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 21/09/2026
5/5 Volodymyr Kuleshov of @cornelluniversity.bsky.social spoke at the Simons Institute workshop on Diffusion Generative Modeling: Progress and Next Steps. Video: simons.berkeley.edu/talks/volody....
simons.berkeley.edu
How to Build a Modern Diffusion Language Model
This talk introduces diffusion for language and the research advances that underlie modern diffusion LLMs. We describe the building blocks of today's leading open-source models, starting from simple m...
040
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 21/09/2026
4/5 "To carry [over the pre-training gains] to post-training and inference time scaling...we need to develop language models that are fully parallel both in training and inference," said Volodymyr Kuleshov of @cornelluniversity.bsky.social at the Simons Institute. Diffusion LMs are an option.
110
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 21/09/2026
3/5 "Both of these techniques are heavily bottlenecked by our ability to generate from the model quickly, both in RL and inference time scaling...the current algorithm used for generation is inherently sequential," said Volodymyr Kuleshov of @cornelluniversity.bsky.social at the Simons Institute
110
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 21/09/2026
2/5 But, "most of the gains [in language modeling performance in the last two years] have come from scaling post-training, especially using reinforcement learning as well as inference time scaling," said Volodymyr Kuleshov of @cornelluniversity.bsky.social at the Simons Institute
110
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 21/09/2026
1/5 The case for diffusion language models: "A lot of the [early] gains in language modeling performance have come from scaling pre-training...[the training algorithm] was designed to be very parallelizable across GPUs": Volodymyr Kuleshov of @cornelluniversity.bsky.social at the Simons Institute
1101
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 31/08/2026
"Of course, we also have failures," said Michal Irani (@weizmann.ac.il), at the Simons Institute workshop on Topics in Intelligence: World Models and Social Reasoning. For e.g., the fMRI scan of a person viewing a cat became a bear. Video: simons.berkeley.edu/talks/michal... arxiv.org/abs/2510.25976
000
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 31/08/2026
4/5 Here are some examples of successful reconstructions of the images observed by a person, using only the fMRI images. For each pair, left is the original image, right is the reconstruction from the fMRI scan. Paper: arxiv.org/abs/2510.25976 Roman Beliy et al.
100
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 31/08/2026
3/5 A diffusion model uses both the initial guess & the CLIP embedding to create high-level images that are semantically meaningful and consistent with the global layout of the original image, said Michal Irani (@weizmann.ac.il, at the Simons Institute. Paper: arxiv.org/abs/2510.25976
100
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 31/08/2026
2/5 Given an fMRI image, a high-level semantic decoder learns to predict the CLIP embedding of the image the person saw. A low-level structural decoder learns to predict crude images from the fMRI image, forming an initial guess of the image's global layout, said Michal Irani, @weizmann.ac.il.
100
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 31/08/2026
1/5 Imagine taking an fMRI scan of the brain of a person viewing an image, and reconstructing what the person saw from the fMRI alone. That's Brain-IT. "This is state-of-the-art image decoding from fMRI," said Michal Irani (@weizmann.ac.il), at the Simons Institute.
111
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 25/08/2026
3/3 "This is part of a larger...collaborative effort to build foundation models," said Eva Dyer of @upenn.edu at the Simons Institute. Such models would unify diverse neural data with varying temporospatial resolutions, from multiple species and tasks. Video: simons.berkeley.edu/talks/eva-dy...
010
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 25/08/2026
2/3 "We've been excited by the idea of taking...fragmented neural datasets & putting them into one unified model...[that] is greater than the sum of its parts," said Eva Dyer of @upenn.edu at the Simons Institute workshop on Topics in Intelligence: World Models and Social Reasoning
110
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 25/08/2026
1/3 From task-specific neural data to foundation models. In neuroscience, studying the neural activity of some behavior in an animal only gives us "a snapshot of the full activity that might be present across the brain," said Eva Dyer of @upenn.edu at the Simons Institute.
140
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 20/08/2026
During interactions with others, "there are synchronizations that appear between two nervous systems...we have this phenomenon within brains. This [also] happens between agents," said Guillaume Dumas, @introspection.bsky.social, of @mila-quebec.bsky.social at the Simons Institute.
041
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 20/08/2026
"When we are interacting with others there is a dynamics that transcends the...people that are participating; these dynamics in return shape the people," said Guillaume Dumas, @introspection.bsky.social, of @mila-quebec.bsky.social at the Simons Institute. Video: simons.berkeley.edu/talks/guilla...
131
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 20/08/2026
The Dark Matter of Neuroscience. For ages, "neuroscience didn't have much to say about two people...in a dynamical interaction," said Guillaume Dumas, @introspection.bsky.social, of @mila-quebec.bsky.social at the Simons Institute workshop on Topics in Intelligence: World Models and Social Reasoning
162
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 18/08/2026
"What we foresee in...cognitive science is that we move progressively from 'others' as a problem to 'others' as a form of affordances for interactions, enabling things that we aren't able to do on our own." said @introspection.bsky.social of @mila-quebec.bsky.social, at the Simons Institute.
012
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 18/08/2026
"The question of 'other' and social cognition is kind of central in cognitive science," said Guillaume Dumas, @introspection.bsky.social, of @mila-quebec.bsky.social quebec.bsky.social, at the Simons Institute. Video: simons.berkeley.edu/talks/guilla...
112
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 18/08/2026
The African philosophy Ubuntu — "I am because you are" — inspires Guillaume Dumas (@introspection.bsky.social) of @mila-quebec.bsky.social, in his research on social cognition and interaction. Dumas spoke at the Simons Institute workshop on Topics in Intelligence: World Models and Social Reasoning
112
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 12/08/2026
3/3 Paul Liang of @mit.edu spoke of his lab's work developing socially-intelligent AI, focusing on modeling touch, olfaction, and the internal states that drive human behaviors, at the Simons Institute. Video: simons.berkeley.edu/talks/paul-l...
000
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 12/08/2026
2/3 Such an AI would be "capable of understanding the environment, understanding [and] interacting with people, eventually building towards [a] social world model," said Paul Liang of @mit.edu at the Simons Institute workshop on Topics in Intelligence: World Models and Social Reasoning
110
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 12/08/2026
1/3 For AIs to be socially intelligent, they'll have to learn internal "social world models" of people & social interactions, not just physical world models, said Paul Liang of @mit.edu at the Simons Institute workshop on Topics in Intelligence: World Models and Social Reasoning
100
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 11/08/2026
The schedule of talks for our Aug. 24–28 ICM Satellite Conference on Spectral Theory, High-Dimensional Expansion, and Pseudorandomness is now online: simons.berkeley.edu/workshops/ic...
062
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 10/08/2026
4/4 "The second key ingredient to [True] AI is not having an objective, [but] to have an emergent objective. To have [an] intrinsic motivation, where...you just want to learn about the world and to discover better data," said Alyosha Efros at the Simons Institute simons.berkeley.edu/talks/alyosh...
000
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 10/08/2026
3/4 "Will this scraped data give us next Borges or the next Bach. I don't think so. At least, I hope not," said @ucberkeleyofficial.bsky.social's Alyosha Efros at the Simons Institute workshop on Topics in Intelligence: World Models and Social Reasoning.
110
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 10/08/2026
2/4 The first ingredient for True AI is data. But is it enough to scrape data and distill 2000 years of human written knowledge into a model? No, said @ucberkeleyofficial.bsky.social's Alyosha Efros, "There is something inferior about stealing knowledge instead of generating it yourself."
120
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 10/08/2026
1/4 "AI is not when computer can write poetry. AI is when computer will *want* to write poetry." @ucberkeleyofficial.bsky.social's Alyosha Efros quoted a friend, when talking about the nature of "True AI," at the Simons Institute workshop on Topics in Intelligence: World Models and Social Reasoning
710
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 06/08/2026
3/3 Continual Learning. "Even people who started the hyper-scaling trend have been talking about continual learning, which is not there in current day systems. Once the system [is] public, it does not learn any more," said UC Berkeley's Jitendra Malik simons.berkeley.edu/talks/jitend...
010
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 06/08/2026
2/3 The Era of Experience: UC Berkeley's Jitendra Malik spoke of Rich Sutton's argument that scaling up AI models "is not autonomous learning, [it's] just memorized regurgitation of some form." Also, there's work on World Models / Dynamics Models as an alternative to scaling up current AIs
110
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 06/08/2026
1/3 AI and Its Discontents: @ucberkeleyofficial.bsky.social's Jitendra Malik talked of the "rumblings of dissent,", when it comes to the current approach of scaling up frontier multimodal models, at the Simons Institute workshop on Topics in Intelligence: World Models and Social Reasoning
182
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 04/08/2026
This week at the Simons Institute, a workshop on Diffusion Generative Modeling: Progress and Next Steps simons.berkeley.edu/workshops/di...
simons.berkeley.edu
Diffusion Generative Modeling: Progress and Next Steps
Diffusion models are now the de facto approach to generative modeling across a wide range of data modalities including images, audio, videos, and visuomotor policies. In recent years, there has been a...
061
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 04/08/2026
Simons Institute Director Venkat Guruswami has been honored this summer with STOC and CCC 2026 Test of Time Awards for papers that transformed coding theory and pseudorandomness, and emerged from a common algebraic core. simons.berkeley.edu/news/guruswa...
simons.berkeley.edu
Guruswami Receives Test of Time Awards at STOC and CCC 2026
We’re delighted to share that Simons Institute Director Venkatesan Guruswami has been honored this summer with two 2026 Test of Time Awards for papers that transformed coding theory and pseudorandomne...
0151
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 03/08/2026
"Robotics is behind [NLP] and computer vision," because the main methods for collecting training data — using teleoperations, human videos and simulations — are all "unsatisfactory," said @ucberkeleyofficial.bsky.social's Jitendra Malik at the Simons Institute. simons.berkeley.edu/talks/jitend...
010
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 03/08/2026
In robotics, "there are problems to be solved which are to do with shortage of data and only when we solve them will we solve robotics," said @ucberkeleyofficial.bsky.social's Jitendra Malik at the Simons Institute's workshop on Topics in Intelligence: World Models and Social Reasoning.
130
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 23/07/2026
Congratulations to all the #ICM2026 Fields Medal and Abacus Medal recipients! And special kudos to theoretical computer scientist Shayan Oveis Gharan, who received the Abacus Medal, and has been a program organizer and frequent long-term visitor at the Institute over the years.
060
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 20/07/2026
Nikhil Srivastava is now the Simons Institute’s Interim Associate Director! He succeeds Sampath Kannan, who returns to UPenn this fall. Taking Nikhil’s place as Senior Scientist is Berkeley Statistics and EECS faculty member @jasondeanlee.bsky.social. simons.berkeley.edu/news/faculty...
simons.berkeley.edu
Faculty Leadership Team Update
As of July 1, our friend and colleague Nikhil Srivastava is the Simons Institute’s Interim Associate Director. He succeeds Sampath Kannan, who returns to UPenn this fall, after two highly successful y...
093
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 09/07/2026
Thursday and Friday this week: a workshop on Quantum Circuits and Algorithms for Cryptography simons.berkeley.edu/workshops/qu...
050
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 09/07/2026
ICM 2026 Satellite Conference: Spectral Theory, High-Dimensional Expansion, and Pseudorandomness August 24–28 at the Simons Institute Join us! simons.berkeley.edu/workshops/ic...
062
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 08/07/2026
Join us! simons.berkeley.edu/events/encry...
010
Simons Institute for the Theory of Computing @simonsinstitute.bsky.social · 02/07/2026
From our friends at Berkeley's Center for Responsible, Decentralized Intelligence: Agentic AI Summit, August 1–2 Early bird pricing ends July 5 rdi.berkeley.edu/events/agent...
rdi.berkeley.edu
Agentic AI Summit 2026
Join us August 1–2, 2026 on the UC Berkeley campus for the largest summit dedicated to Agentic AI — 5,000+ attendees, world-class speakers, and a global livestream.
120