Jun Li

It's a foundation model, which is a general purpose model trained on broad data that can be adapted for many tests.

AI2 said the platform is end-to-end from raw data to fine-tuning and deployment with tools for labeling and evaluation.

Business Wire notice adds that the training corpus spans roughly 10 terabytes with examples like wildfire risk mapping and surface water change detection.

The models and pipeline aim to help governments, NGOs, and local teams who lack geospatial AI resources.

The differentiator here is openness and reproducibility across the entire stack from models to data pipelines, which AI2 has emphasized across its OLMO efforts.

They've come out with a bunch of different projects all under this same project banner.

While their peers have built Earth AI models before, but full openness pipelines with productive deployment tools are rarer.

The question is whether open infrastructure can lower the cost time to move from research to field decisions like fire mitigation and illegal fishing detection without sacrificing quality and safety.

Early users will test if the package stack reduces specialized engineering effort while meeting accuracy targets in the wild.

So the thing that drew me to this story is

Overall, the AI2 organization has been really focused on the AI for good kind of thing.

And it's been either biological or ecological in focus.

And this project is definitely ecological in focus.

And what's really nice about this is, like I said earlier, it is an open source platform.

So right now it's in beta and you get beta access.

You got to sign up for it.

They'll evaluate your access and usage and your project that you want to use it for, your purpose you want to use it for.

But essentially it's for projects and researchers and scientists who want to use this information or this data set, essentially this platform for lots of early warning

kind of things.

And a 10 terabyte data set is a big data set.