I’ve lately been asking my AIs of choice (Gemini and Claude) to help with small decisions or doubts I have. On one hand, I fear that by unloading the cognitive weight of decision-making, I might be making myself "weaker", much like how taking the car for even the shortest trips doesn't exactly help you develop cardiovascular strength. On the other hand, I suspect I often do it just to give myself more arguments for what I already believe, such as: "Estimate which hardware setup uses more power with this data".
For now, I only use it for technical (and unimportant) topics. I refuse to use it for other parts of my social or political life, as I believe it could act as an extremely appealing echo chamber.
I’ve also been using AI to refine my ideas, asking questions as I’m working through a specific problem in my mind. It helps me find new angles and discard bad ideas faster.
Another (more practical) use case was providing the following prompt and letting it reorganize my FreshRSS instance.
Access my FreshRSS server using the GReader API at $FRESHRSS_URL. The user is stored in the environment variable $FRESHRSS_USER and the password is stored in the environment variable $FRESHRSS_PASS. Return the subscriptions I have and the section/folder in which each one is located. How could I better organize the different feeds? Which additional tags would you create?
After working through some small decisions with Claude, it seems to have done a pretty good job. I suppose I'll sense the real results in a few weeks.
I should probably expect it by now, but it’s still crazy to me that it can successfully find the schema of an API and interact with it with almost no errors, even when the requested action requires multiple API requests.