Autonomous AI agents could change the world, but what do they actually do well?

The two most important problems (at least how I am thinking about them currently), are:

Finding a rigorous scientific framework for how different agent skills, personalities, and instructions combine to be most capable for different problems (think of this as social management science for AI agents) .

Figuring out how you formally validate and verify... See more

From chatbot to agent-coworker

What could ambitious unhobbling over the coming years look like? The way I think about it, there are three key ingredients:

1. Solving the “onboarding problem”

GPT-4 has the raw smarts to do a decent chunk of many people’s jobs, but it’s sort of like a smart new hire that just showed up 5 minutes ago: it doesn’t have a... See more

An LLM’s abilities are amplified when combined using emerging agent frameworks like BabyAGI, AutoGPT and ChatArena. These frameworks break down workflows into a series of discrete roles each handled by AI agents, which are given a defined objective and work together to automate a complex task.

Initially, this will take the form of human-AI collabora... See more

And these AIs will be able to take actions. You will just give it a general, high-level goal and it will use all the tools it has to act on that. They’ll talk to other people, talk to other AIs. This is what we’re going to do with Pi.

Although the practical success of neural networks is still undeniable, it must be observed that their most powerful applications are in domains where rules are set in advance and don’t change over time — where the goals are clearly defined. The way you win a game of chess or Go is fixed and unambiguous. Protein folding is constrained by the laws of... See more

Although the practical success of neural networks is still undeniable, it must be observed that their most powerful applications are in domains where rules are set in advance and don’t change over time — where the goals are clearly defined. The way you win a game of chess or Go is fixed and unambiguous. Protein folding is constrained by the laws of... See more

In a pre-AI world, in order to build an application, you needed to reduce your idea to a process—a set of rules by which your software would operate to accomplish the goal you set. Sometimes this was easy; for example, a customer relationship manager like Salesforce is naturally reducible to rules.

In a post-AI world, you can build applications for ... See more