OpenAI executive Colin Jarvis explores how AI systems will become more powerful through the ability to reason. “We think 2025 is the year of AI agents”, he says. “You can think of them as almost the assistant that you can talk to.”
Jarvis begins by explaining OpenAI’s three-tier structure: research pushing the frontiers of AI, an applied team productizing innovations, and the deployment team that tests these innovations in real-world scenarios.
“What we don’t want is to try and develop AGI in a lab”, Jarvis emphasizes, referring to Artificial General Intelligence. “We need to kind of battle test it, put it in people’s hands and make sure that it actually has utility for people in their day-to-day lives.”
In his presentation, Jarvis showcases the power of multimodal capabilities combined with reasoning. In one demonstration, an AI assistant engages in natural conversation while simultaneously analyzing a complex aerodynamic diagram.
Thanks to reasoning, “we can interact with them in very natural ways that are very different from how we used to use AI before”, Jarvis notes. “This is where we see a lot of the value potential for agents.”
Watch the video for further examples and to go deeper into the technological advances making reasoning in AI possible.
