Diary 2025-01-30

Voice conversation with AI

Now that Devin has been privatized, we will proceed with the source code organization of the public opinion map and the analysis of the MVID dataset.

Each is left to Devin.

It used to be best practice to take small steps with respect to tasks that were too big, but that may no longer be the case.

It would be better to break down the tasks and consider where AI can move them.

In many cases, just breaking down the task is not in a form that can be delegated to AI.

The reason for this is that, at present, systems have not been built with the assumption that AI will be entrusted with tasks.

In this light, the task that should be given the highest priority is the creation of such a system.

(4o) To design a system for creating a mechanism to delegate tasks to AI, it is important to first clarify the specific requirements of the task and identify which parts of the task AI can handle. It is also important to design reusable modules and unify the format of data input and output. If necessary, it is a good idea to maintain documentation and make it easy for the AI to reference.

I agree. It is important to reduce friction between input and output.

When we see the downside of return on our efforts in a task that involves uncertainty, we tend to hold back, thinking that our efforts may have been in vain. However, since high uncertainty means high upside, it makes sense to make it possible to take on the challenge at a low cost.

It is a kind of meta-layered programming to realize a goal by combining AI, programs, and other components with different performance and behavior.

Right knee tends to externally rotate

Don't forget to make a document.

---