Teaching AI to understand data
Our world is teeming with information, but right now, only experts can access the knowledge behind that information. We envision a future where complex data analysis is as simple as asking a question.
Vision
We are building a more data-driven world.
In the future, an AI data analyst is universally available. This will drive faster scientific discovery, combat misinformation, improve public policy, promote personal and public health, and strengthen economies.
What We're Building
1
Repository of tabular data
Increasingly large training data sets are a key driver of the rapid increase in AI intelligence. We are building a repository of tabular data with sufficient volume and diversity to train a foundation model. This data set will be similar in scope to The Pile and LAION-5B.
Multi-modal text and data foundation model
Foundation models like GPT-4 have demonstrated that AI can achieve human level-competency in tasks across multiple domains. We are building a multi-modal foundation model for text and data that will understand and manipulate data as well as human data analysts.
2
3
Natural language interfaces to data systems
Our AI data analyst will engage with humans as a collaborator. It will need to clarify and select approaches amidst ambiguity, explain how it arrived at its answers, and adjust it's approach with feedback. All of this will be done in natural language.
