Part 3/9:
Another compelling upgrade is the models’ proficiency in tool utilization. Through reinforcement learning, the new systems are trained to understand not only how to use various tools but also when it is appropriate to deploy them. This enhancement significantly broadens the functional capabilities of AI agents. OpenAI’s President, Greg Brockman, shared insights on the models' ability to make upwards of 600 tool calls during complex problem-solving tasks, hinting at a potential leap towards fully autonomous agents.