The world's leading venue for collaborative research in theoretical computer science. Follow us at YouTube.com/SimonsInstitute.
Berkeley, CA
- Tweets 1,486
- Following 351
- Followers 10,001
- Likes 812
1/4 Are diffusion language models ready for the real world? Not quite. Today's diffusion language models are missing some key ingredients, said @volokuleshov of @Cornell, at the Simons Institute workshop on Diffusion Generative Modeling: Progress and Next Steps
1
2
9
1,222
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 @volokuleshov of @Cornell at the Simons Institute.
1
298
4/4 "This allows you to interpolate between auto-regression and diffusion," said @volokuleshov of @Cornell at the Simons Institute workshop on Diffusion Generative Modeling: Progress and Next Steps. Video: simons.berkeley.edu/talks/vo…
191
Join us Tuesday, 9/29 for the first Richard M. Karp Distinguished Lecture of this academic year, featuring @BooleanAnalysis.
simons.berkeley.edu/events/m…
7
28
2,270
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/particip…
2
15
97
15,352
Simons Institute for the Theory of Computing retweeted
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," said @volokuleshov of @Cornell at the Simons Institute
1
2
17
3,639
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," said @volokuleshov of @Cornell at the Simons Institute
1
2
17
3,639
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 @volokuleshov of @Cornell at the Simons Institute. Diffusion language models are an option.
1
1
2
370
5/5 Volodymyr Kuleshov, @volokuleshov, of @Cornell spoke at the Simons Institute workshop on Diffusion Generative Modeling: Progress and Next Steps. Video: simons.berkeley.edu/talks/vo….
1
4
341
Simons Institute for the Theory of Computing retweeted
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 (@WeizmannScience), at the Simons Institute.
1
3
18
2,071
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 (@WeizmannScience), at the Simons Institute.
1
3
18
2,071
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.
1
1
1
278
"Of course, we also have failures," said Michal Irani, 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/mi…
Paper: arxiv.org/abs/2510.25976
1
268
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 @Penn at the Simons Institute.
2
3
16
1,558
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 @Penn at the Simons Institute workshop on Topics in Intelligence: World Models and Social Reasoning
1
2
682
3/3 "This is part of a larger...collaborative effort to build foundation models," said Eva Dyer of @Penn 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/ev…
3
361