I’m cautiously optimistic about a new research project I started this week. This doesn’t mean I’m abandoning all my work from the past year, but I’m shifting my priorities while the line is still hot.

There’s 2 reasons why I’m excited:

• First, I think this is an important problem. If the system works, then I’ve made progress towards addressing an important problem. This is a great motivator. As Dr. Richard Hamming said, “Once you get your courage up and believe that you can do important problems, then you can.”

• Second, the work involved to build the system is well-suited to my strengths. When you do research in this field, you often have to write a lot of complex code just to run a simple experiment that might not go well, then you have to start over again. The code was just fuel you had to burn to earn new information. In contrast, writing the code for this project is the goal itself. This doesn’t mean that success is guaranteed, otherwise it wouldn’t be an important problem, but that’s something I can deal with. I might just enjoy it.

Now, I won’t describe the full details right now because 1) I’m exhausted and 2) I want to keep some mystique around while it’s still fun. But, I’ll leave this crumb because I thought it was a neat analogy: if an LLM workflow is a program, then a context window is a process, an agent is a thread, and I want to build the compiler.