A nice work from @dmorris! See his LK post. 

"Lots of information in the conservation literature is represented in maps (e.g., the exact survey area associated with a population estimate), so if LLMs are going to be useful for discovery and review in this area, they're going to have to get good at understanding not only the content of maps, but the spatial *context* of maps. That is, given a map, what area on the actual Earth does this map represent?

To help assess how LLMs are doing at this task, and more generally to rile up some discussion about this problem that's relatively unique to the conservation/ecology space, they released a small benchmark dataset:"

Read "GitHub - google-research/ecology-georeferencing" on GitHub

GitHub - google-research/ecology-georeferencing

Contribute to google-research/ecology-georeferencing development by creating an account on GitHub.

GitHub