A new challenge to find lost ancient sites in the Amazon using AI.
Use OpenAI’s latest models (o3, o4-mini, GPT-4.1) with satellite images and other open data.
Best teams get to present their findings and can win prizes.
Learn more:
Looks interesting! If i have time i definitely want to try it
This is such an awesome challenge! As someone who uses ChatGPT primarily for historical and scientific research with a digital humanities lens, I can’t wait to see what ideas come out of this. These tools have worked wonders for synthesizing billions of years of cosmic history and millions of years of astronomical and geologic processes 
I could only imagine what is possible with the right combination of primary sources, GIS data, and OpenAI models for archaeology. The dreams from the era of knowledge-based expert systems can finally be achieved. Best of luck to everyone taking part in the challenge! 
1st place will receive $250,000 in a cash/API credits mix as well as funding to continue their work in collaboration with archaeologists
Looks like they found a cheap alternative to “we’ll train a custom AI for you for $2-$3 million”. “Write thousands of dollars of data retrieval and synthesis code for us for $100 in credits”. Just outsource speculative geographic information systems applications, and have others do uncompensated work – while spending on the API. (“and be sure to shout out on social media”)
I suppose that if A-Z starts with Amazon, there’s a Biology or such coming next.
They just completely blocked fine-tuning gpt-4.1-mini and nano on images instead of letting the smaller fine tuned models be used for vision. And those are the ones that give “patches” vision algorithm and higher resolutions. Ensuring there is no cheap vision for this challenge.
Multispectral remote sensing for a small bombarded island can rack up hundreds of gigabytes of ArcGIS files.
If you’re like me and want to do this but have zero knowledge, here’s the study guide I’ll be following, created using Deep Research: ChatGPT - Amazonian Discovery Crash Course
It isn’t a comprehensive resource, but it’s a good starting hub for noobs starting from nothing. Coding knowledge required.
I was waiting to hear your criticisms on this, haha.
Platform-wise, we’re really all on our own with this, and basically doing this only for fun and for science. All we’ve got to work with is a PDF with some links and example submissions. But OpenAI…
Why not post-train a vision model to only respond to geo data, and give us half off on that until submissions are up?
Why not implement vision fine-tuning for the cheaper patches models so we can actually afford to run inference on large areas of the Amazon on our own dime?
I understand that OpenAI is paying out for this, but no one here is making top 3. I really wish we had more motivation than “it would be cool etc etc.” It just doesn’t feel like we have any real support.
Oh, and has anyone read the book or seen the movie Holes? If so, you’ll realize that there’s an eerie concern looming above the fact that we really don’t know where this data is going, what it’s for, or how it will be used.
Yep, it is "thanks for the free stuff, but we’ll be picking who is good for optics.
it’s merging the data plates and analyzing
and perhaps trying to find variations in tree placement with certain means over time…
geography shape and height patterns…
it could be done,
the map is a chore to make tho…
needle in a haystack with a 157 billion dollar thought bazooka…
very interesting stuff still, after the confetti is removed by your insights