Inside the Development of the Most Potent Open Source AI Model in the World

About a dozen engineers and executives from Databricks, a data science and artificial intelligence startup, convened this past Monday in conference rooms connected by Zoom to discuss whether or not they had been successful in creating a state-of-the-art language model for AI. The group has trained DBRX, a sizable language model with a design akin to OpenAI's ChatGPT, for months and roughly $10 million. However, they would have to wait for the outcomes of the creation's last testing to find out just how strong it was. Eventually, Jonathan Frankle, chief neural network architect at Databricks and head of the team that created DBRX, told the group, "We've surpassed everything." The group responded with whoops, cheers, and emojis for applause. Frankle doesn't typically drink coffee, but she was sipping on an iced latte.