Agentic Horizon Today’s best machine learning models have a high degree of intelligence, but very little agency. Reasoning models like DeepSeek R1 and o3 can solve tasks at an increasingly high level of difficulty across a wide variety of domains, often far exceeding the capability of any single human. But they can’t break down a fairly unambiguous task that would require a year of dedicated work, sit down, and deliver a solution. (An example of such a task: Write me a performant simulato...
