# OpenAI to Z Challenge

## Метаданные

- **Канал:** OpenAI
- **YouTube:** https://www.youtube.com/watch?v=0Ka-Je6mIs8
- **Дата:** 28.08.2025
- **Длительность:** 9:45
- **Просмотры:** 21,872

## Описание

10,000+ people joined the OpenAI to Z Challenge to explore how AI can push the archaeological frontier in the Amazon.

The winner is Team Black Bean, which used deep learning on public LiDAR + satellite data to build maps that surface what's under rainforest canopies.

➡️ Black Bean’s submission: https://www.kaggle.com/competitions/openai-to-z-challenge/writeups/amazon-archeological-site-discovery-a-deep-learnin

## Содержание

### [0:00](https://www.youtube.com/watch?v=0Ka-Je6mIs8) Segment 1 (00:00 - 05:00)

I'm so grateful to be chosen as one of the five finalists. — I want to introduce you to AKOS, a scalable system that I built. We have integrated open AI into the application to print an intelligence into the system. — There are a lot of different types of earthworks and they each have their own characteristics. — We decide to use deep learning to train classifiers on lighter data and satellite images to classify each segment of the Amazon forest. We divided the region tile by tile for 3x3 km and for each centroid of these tiles we run the model again and we extracted all the parameters for this prediction and detection models. — We made some configuration changes over the custom pro and you can see the noise has been reduced and some of the features have been clearly visible. There are around 100 plus potential sites that we have found. We created an interactive website and when the user click on a spot they can dive into the details. We went to GPT GP based tri is done and we told GPT to act as an archaeologist with years of experience and give a final report. — The deep learning approach is a scalpable enough to scan the whole Amazon rain forest in a reasonable duration of time. this approach can really work and then can actually help the archaeologist to do the discovery more efficiently. — The energy and enthusiasm of the participants has been extraordinary. Everyone who's made the finals has something really interesting and unique. — Just thinking about all the big LAR projects and that could be like super groundbreaking. — This is so new and so different and it could be something really powerful. Hi, Philip. Hello. — Hi. Congratulations team black bean and welcome. — Thank you. I feel like you've inspired many different folks around the world and we're just very excited to share you guys as winners and the work that you've done to the world. Feel free to take a minute to introduce each other um and tell us why you named the team Black Beam. — And I'm Yao and I'm current I have a year working experience as software engineer and currently working in Manta. Uh my name is Inia and I'm currently a software engineer in Amazon. — I'm Ya and now I just graduated from Nan Technological University. — And for the uh team name Blackbean. Blackbean is the family dog uh who passed away a few months ago at 16 years old. So we decided to use this name to remember him. — Well, Black is going to be shown to the world. So definitely made him proud. Uh was how did you come across this challenge? — Yeah, actually I found it first. So last year I took gap year uh I quit my current job and uh starting learning um machine learning and basic some basics and also some like deep learning techniques. Um and uh and you know cargo is really famous uh for like holding the data scientist computation. So in the in May this year when I like casually browse the CO I found this competition I found this super exciting. Um so and I actually have the interest going and then I contact Yan uh who happen to have the availability so we form a team and uh start the journey. — Yeah. Y'all told me on my summer vacation so I haven't enough time to do it. I'm also uh I'm in charge in some Asian civilization. So I'm really excited to take part in — talk us through some of your findings and how you leveraged open AI's models to approach the challenge. — Uh we actually found the deep learning is really good approach to help archaeologist to discovering archaeological site especially on Amazon area. So um that's why I we we decided to uh go for this approach and uh and in the end we finally like gather all the data and do all the training and post-processing and uh and we finally drill down to some like a small number of results. This would this approach can really work and then can actually help the archaeologist to do the discovery more efficiently for the open air model. I think uh it's not just a question ask and uh answer chat boat. I think this is more like a collaborator for many times. I ask him to uh what to do next. Uh he can memorize the whole

### [5:00](https://www.youtube.com/watch?v=0Ka-Je6mIs8&t=300s) Segment 2 (05:00 - 09:00)

dialogue over months or a year. So he know the whole structure of our project. uh for each point he will give me some uh several ways for me to choose. We discuss about the stress the weakness and finally choose the uh the solution. I think I even discuss with Chivity more times than my teammate. So yeah, I think it's a quite great collaborator. Yeah, — that's incredible to hear. And out of curiosity, what was your biggest wow moment during this challenge? — Yeah, the big biggest wow moment definitely is like when we get the result from uh from our model and after like all the post-processing we find something we find like a list of like u potential discover site and like when we do the manual analysis we found them uh actually really has some potential like based on our shadow archaeological knowledge and also some common senses. So that's a moment that we feels like actually it can really work. It can really extract characteristic of the training material and then to apply it in the model and then it can really do the classification like on the segments of the Amazon rainforest. So we really found that our approach can really work can really help archologists. — And what would be your advice for how other people can use LMS for research and discovery? — CHBT it's a good feature. uh you know we have firstly we have not know no knowledge about the archeologic archeologist in aman basin uh it teach us only one week and we become some part of the arologist I think maybe uh and then it's a collaborator as I said uh it's maybe not the best uh best model we use you know charg features it's only can understanding some picture and give me some dialogue. So we make it as a picture description I think but not the uh C model you know I want to say more about that I think that that's I think that the one like uh CH is really good at is like to do a summarization so uh in the realization part we actually like summarize each like potential spots like uh to provide a very long text I think that's a very good way to help archaeologists to understand like why this uh spot is chosen by our model and also so we think that uh chibi is a is a really good summarization tool and can really can provide a very good way to show the work maybe show some work with really a deep domain knowledge like to the to the broader audience. So I think that this is really incredible. Um well before we wrap up we actually want to say a special thank you to the incredible support from the broad archaeological community um as well as our friends Sarah Parkak and Chris Fisher who you all had met through the judging panel. Uh we'd love to hear about your plans on how you continue this work and how you intend to collaborate with other finalists and participants to continue pursuing this effort. I think we can help them to public to make our work public and so we can get feedback from them and it also at the same time inspire them to do more. I think there are still rooms of improvement for our approach. So we probably need to like spend some time to really to improve our approach and then maybe we can we can share this work with like even broader community and this work might be utilized to uh even different kinds of archeological research or even inspired like in other fields. — Before we let you guys go is there anything else you want to share with the world or with black bean? — Yeah. Um I actually want to uh say thank you to open eye uh to organizer Philillip and uh and also to all the judges in the panel. Uh I we really appreciate that uh you provide this opportunity uh for us to do this very exciting and interesting project and we really want to uh continue this uh this work and uh to hopefully can benefit uh the research on side discovery research on Amazon rainforest and even like a broader scope. — Well, thank you team and uh and massive congratulations. — Thank you.

---
*Источник: https://ekstraktznaniy.ru/video/11275*