Since I started working on AI in 2016 I’ve developed a clear motivation: SUPER AUTOMATE EVERYTHING.
Now that has new meaning. Talking to my colleague Sewoong Oh who was working on DataComp, I had a realization: the process of building state of the art AI models itself is a process that is quite manual and slow. In DataComp for example, while the paper has 50 or so authors, many of these talented AI researchers had to spend a lot of their time working on highly repetitive manual data processing tasks. This led me to examine further the opportunities for using AI to build AI.
Since I started this line of thought and research, we have seen a number of amazing projects and breakthroughs in this direction. In this series of posts I’ll be making as part of a “Season of Reasoning” I hope to highlight some of these breakthroughs, how that is relevant to our research at Sentient and the path to AGI.
To get us started here’s the first post I wrote on AgentInstruct: Using AI to build AI: AgentInstruct in Depth Summary