The Early Days of AI2

We're talking about thousands of pages of documents that normally a lawyer would need to review, get an eyes on each page and find out what different terms are. And with our prototype back then, our goal was how can we quickly go from here's a pile of documents to here are very high quality AI models that can extract this information reliably? We were highly inspired by the deep dive project, which eventually became the snorkel project and looking at their approach of weak supervision. The other cool thing to recognize is in 2018, there was a groundbreaking paper that came out of AI2 called the Elmo paper. Google built on top of it and cheekily called it Burt.