Ron Shprints
122 posts
There's something refreshing about reading a pre-2023 paper and being absolutely certain that it was all human generated. As much as I will do research on LLMs, I will never use them to write a word of my papers.
English
Ron Shprints retweetledi
Very excited about this work! Language modeling via flow maps allows us to rethink how to build language models that do latent reasoning and fast generation!
Glad to play my small part in a wonderful effort with @PPotaptchik, @json_yim, @adhisarav, Eric Vanden-Eijnden, @msalbergo!
Michael Albergo@msalbergo
New paper! Presenting Discrete Flow Maps: paper: arxiv.org/abs/2604.09784 blog: malbergo.me/discrete-flow-… A laughable problem for me these days is that @nmboffi and I share a research brain, and we have had, time and again, a conversation that ends with “ha so I guess we’re
English
Ron Shprints retweetledi
Arriving in Rio 🇧🇷for #ICLR! I will present GLASS Flows + Diamond Maps in the coming days - starting today with GLASS Flows Oral at 11:30 AM! I am looking forward to meeting new people! Feel free to reach out!
1/3
Peter Holderrieth@peholderrieth
We release Diamond Maps💎 unlocking accurate and efficient guidance for diffusion models. Our experiments show that our methods scale incredibly well. Excited to see what people will build with this! Accurate guidance has been a notoriously hard problem, but in this work, we’re
English
RT @peholderrieth: We release Diamond Maps💎 unlocking accurate and efficient guidance for diffusion models. Our experiments show that our m…
English
@andrewgwils I'm sad to admit, this might not even be possible for me.
English
I did an experiment a couple years ago where I completely unplugged from all computer technology for two weeks. After a couple days of fomo and withdrawal, I've never felt better or more deeply focused, since the 1990s.
English
Ron Shprints retweetledi
We are also releasing self-contained lecture notes that explain flow matching and diffusion models from scratch. This goes from "zero" to the state-of-the-art in modern Generative AI.
📖 Read the notes here: arxiv.org/abs/2506.02070
Joint work with @EErives40101.
Peter Holderrieth@peholderrieth
🚀MIT Flow Matching and Diffusion Lecture 2026 Released (diffusion.csail.mit.edu)! We just released our new MIT 2026 course on flow matching and diffusion models! We teach the full stack of modern AI image, video, protein generators - theory and practice. We include: 📺 Videos: Step-by-step derivations. 📝 Notes: Mathematically self-contained lecture notes 💻 Coding: Hands-on exercises for every component We fully improved last years’ iteration and added new topics: latent spaces, diffusion transformers, building language models with discrete diffusion models. Everything is available here: diffusion.csail.mit.edu A huge thanks to Tommi Jaakkola for his support in making this class possible and Ashay Athalye (MIT SOUL) for the incredible production! Was fun to do this with @RShprints! #MachineLearning #GenerativeAI #MIT #DiffusionModels #AI
English
Recordings have just dropped on YouTube! You should definitely check out this amazing resource if you want to learn about flow matching. Peter is the best of the best and he put so much effort into this. It was a pleasure to contribute my small part:)
Peter Holderrieth@peholderrieth
🚀MIT Flow Matching and Diffusion Lecture 2026 Released (diffusion.csail.mit.edu)! We just released our new MIT 2026 course on flow matching and diffusion models! We teach the full stack of modern AI image, video, protein generators - theory and practice. We include: 📺 Videos: Step-by-step derivations. 📝 Notes: Mathematically self-contained lecture notes 💻 Coding: Hands-on exercises for every component We fully improved last years’ iteration and added new topics: latent spaces, diffusion transformers, building language models with discrete diffusion models. Everything is available here: diffusion.csail.mit.edu A huge thanks to Tommi Jaakkola for his support in making this class possible and Ashay Athalye (MIT SOUL) for the incredible production! Was fun to do this with @RShprints! #MachineLearning #GenerativeAI #MIT #DiffusionModels #AI
English
@andrewgwils I started as a chemist. Took a long detour in "pure" AI for a while. Now I partake in the so called "AI4Science" party. Am I a scientist? Or an ML researcher? I wouldn't know. I'm probably neither. But I certainly enjoy both.
English
Presently, many ML researchers dabble in scientific applications, as part of what is called "AI for Science". In the future, domain scientists will become AI scientists, and what it means to be an AI researcher will evolve to be more narrowly focused on foundations.
English
@andrewgwils I'm going to be nerdsniped for this, but when I figured this in high school I was absolutely blown away. For weeks, my go to conversation starter was the microwave story. I just couldn't believe that we managed to put this machine together. Oh god I've just got nerdsniped again.
English
The microwave must have been mindblowing when it came out. Even now, it's so much more futuristic and cool than "cutting edge" AI technology. The most sci-fi device in our lives.
English
Ron Shprints retweetledi
Most people assume you need a massive dataset to distill flow models.
We challenge that.
Is data actually necessary? Or perhaps it is a liability?
Introducing FreeFlow: We achieve SOTA (1.49 FID on ImageNet-512) 1-step image generation without a single data sample.
🧵👇[1/n]
English
"Analyzing NBA player positions and interactions with density-functional fluctuation theory"
nature.com/articles/s4159…
English
Ron Shprints retweetledi
Ron Shprints retweetledi
New work led by @peholderrieth showing how to transform an already trained flow matching model to a stochastic transition/posterior model that can still be sampled via an efficient ODE solver!
Peter Holderrieth@peholderrieth
New work: “GLASS Flows: Transition Sampling for Alignment of Flow and Diffusion Models”. GLASS generates images by sampling stochastic Markov transitions with ODEs - allowing us to boost text-image alignment for large-scale models at inference time! arxiv.org/pdf/2509.25170 [1/7]
English
Ron Shprints retweetledi
New work: “GLASS Flows: Transition Sampling for Alignment of Flow and Diffusion Models”. GLASS generates images by sampling stochastic Markov transitions with ODEs - allowing us to boost text-image alignment for large-scale models at inference time!
arxiv.org/pdf/2509.25170
[1/7]
GIF
English
@TrevorCampbell_ @AllThingsApx Nah it's definitely the Kendall Square Cava
English
What is the AI x Bio equivalent of the Pentagon Pizza Report?
English
Ron Shprints retweetledi
a few months ago we @_DimensionCap quietly funded a team of 3 MIT computer scientists & mathematicians at the forefront of ML & physics+simulation for enzyme design. these. guys. are. cracked
today they're stealthily building the early team. if you're world class, look below 👇
English
@dvruette Wdym by poor shuffling in the data loader? Ordering is similar to the original dataset?
English
issues were numerical stability of loss function (fixed by manual mixed precision) and poor shuffling in the streaming data loader (fixed by dask.dataframe -> df.sample(frac=1.0))
English
at last, it is fixed!
there shall be no more funny-looking loss curves from here on out 🙏
Dimitri von Rütte@dvruette
now we've gone from roller coaster loss to the spikiest loss curve known to man. progress..?
English
Ron Shprints retweetledi
Thanks @MIT_CSAIL for featuring our work!🖊️🎨
Huge thanks to the CSAIL news team for the fun article + video!!
We'll be presenting SketchAgent at #CVPR2025 next week — come say hi if you're curious how LLMs can be used to collaboratively sketch!🖌️
👉 bit.ly/43mTme1
MIT CSAIL@MIT_CSAIL
Sometimes the best way to express an idea is by sketching it out. A system from MIT CSAIL & Stanford captures this iterative process by teaching LLMs to create sequential sketches. It could work w/users to visually communicate concepts: bit.ly/4kfXFhk
English