Project ProtoStar wants publicly funded science to be ready for AI, and you can join the first challenge 🔭

If taxpayers fund research, why is the data behind it so often hard to use? On 29 October Hugging Face and a group of American universities announced Project ProtoStar, an initiative to make publicly funded datasets, models and code easier to discover, reuse and build upon, by people and by AI systems alike. It is not a flashy model launch, but it might matter more in the long run.

Public is not the same as usable

The U.S. federal government invests roughly $190 billion a year in research and development, and agencies like the NIH and NSF increasingly require the results and the underlying data to be shared. But shared files often sit in legacy formats, are scattered across incompatible systems, or come without the documentation, metadata and provenance you need to understand them.

ProtoStar wants to close that gap. Participating organisations sign a memorandum of understanding to release suitable scientific artifacts in AI-ready, machine-actionable formats on open infrastructure. The founding partners are Carnegie Mellon University, Purdue, Georgia Tech, Arizona State University and The George Washington University (working through the Open Forum for AI), plus the Astera Institute, with Hugging Face as infrastructure partner and ORCA as community partner. According to Data Phoenix, the agreements are non-binding and non-exclusive, and each organisation keeps control over what it releases.

The numbers behind the idea

The figures Hugging Face shares show the demand is real. The Hub already hosts more than 17,000 public datasets in core scientific domains such as genomics, protein design, chemistry, climate, Earth observation and astronomy, close to 29,000 if you add medicine and mathematics. Researchers shared more than three times as many new science datasets in the first nine months of 2026 as in the same period of 2025. The 18 repositories from NIH's NCBI have passed 11 million downloads. And my favourite detail: 90% of dataset downloads on Hugging Face come from fully documented datasets. Good documentation is not bureaucracy, it is what makes data travel.

A climate challenge you can actually join

ProtoStar will run a series of open challenges, each hosted by a partner institution. The first, Carnegie Mellon's Climate Risk Intelligence challenge, runs from 2 to 13 November 2026. Teams get historical weather records for 210 U.S. cities up to 2018 and must predict what actually happened in the years after: average annual temperature, total precipitation, the hottest day of the year and cooling degree days. Predictions are compared against simple baselines like historical averages and linear trends, which makes the question honest: does AI really beat the simple approach? Teams also propose a short climate adaptation policy idea.

Why I like this

The challenges are designed to learn what makes reuse possible, including documentation gaps and interoperability problems, and to feed that back into infrastructure and policy. That is the right mindset, and Lorena Barba of George Washington University put it perfectly: "Public access is a good first step. Usability is next." ProtoStar is U.S.-focused for now and its impact depends on how many institutions follow through, but I would love to see this model in Europe too.

Sources

Open data that is actually usable is how small teams get to do big science. I'm curious which forecasts, and which policy ideas, come out of that first challenge.

Reacties

Een reactie posten

Populaire posts van deze blog

Meta just shared the full recipe for a 30B agent model you can run at home 🧠

When Meta released Muse Glimmer in August, the pitch was simple: an open-weight, Apache 2.0 model built for AI agents that run on your own Mac or PC. What was missing was the "how". On 19 October, Meta Superintelligence Labs published the full technical report, and for anyone who cares about distillation and local agents, it is one of the most useful reads in a while. What the model is Muse Glimmer is a dense model of about 29.6 billion parameters, including a 1.8B perception encoder, so it reads text and images and writes text. It has a 128K+ context window and was trained on data from more than a hundred languages. The design goal was strict: everything has to fit in 32 GB of GPU memory, and a smaller 4-bit variant targets 24 GB cards. The parts I found most interesting Dense beat mixture-of-experts here. That surprised me. The team trained scaling ladders for a dense model and two MoE designs with the same memory footprint. Under a fixed memory ceiling and their compu...
Meer lezen

AI that can actually "see" a crystal 🔬

Most of the solid stuff around us, from battery materials to chips, is made of crystals: atoms stacked in neat, repeating 3D patterns. Large language models are great at reading papers about those materials, but they have never really looked at the structures themselves. A team at MIT just showed a clever way around that, and the trick is one I keep coming back to: good synthetic data. What they built The model is called CLaC, short for Contrastive Language-Crystal model, and it is described in a paper in Nature Communications by Yang Jeong Park, Mayank Kumaran, Chia-Wei Hsu, Elsa Olivetti and Ju Li. MIT News covered it on 18 October 2026. If you know CLIP, the model that taught image generators to connect pictures and words, you already get the idea. CLaC has two branches: a graph neural network that reads the 3D crystal structure, and a text encoder that reads a description. Contrastive training pulls matching structure-text pairs together in one shared space and pushes mismatche...
Meer lezen

A 35B model on a Mac mini, using just 2.9 GiB of memory 🚀

There's something special about running a capable language model on hardware you own: no API bill, no data leaving the house, just a small box on your desk doing the thinking. A paper that's currently #2 on Hugging Face's trending list takes that idea further than I expected: it runs a 35-billion-parameter model on a Mac mini at about 20 tokens per second while using under 3 GiB of memory. The trick: leave the experts on the SSD The paper is called The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction , and the system is named Edge0 . It builds on a Mixture-of-Experts (MoE) model, Qwen3.6-35B-A3B. In an MoE, each token only uses a few "experts" per layer (here 4 out of 256 in each of 40 layers), so most of the weights sit idle at any given moment. Edge0 takes that literally. The expert weights (quantized to int4) stay on the SSD and are streamed in only when needed. The clever part is a small prerouter : a tiny trained he...
Meer lezen