LLMs are making me lose my savviness

I find myself using LLMs locally because they are a tool, and like any other tool I've discovered and used over the years, I want to use and master them. However, this usage is rather boring. The allure of LLMs lies in their ability to speed up the design and development process, but it has left me feeling bored and unfulfilled.

Gone are the days when I would spend hours and sleepless nights thinking about problems and solutions, only to rush back to my desk to write and implement them. That thrill of creativity and satisfaction has vanished, replaced by a monotonous cycle of prompting, evaluating, adjusting, and repeating. There's no engineering or challenge left - just the act of prompting and evaluating.

It's like reading a summary or description of a math formula without applying it - quite pointless. I've always been a "maker," deeply involved in understanding topics, spending time on them, and experiencing the satisfaction of creating something meaningful. But now, the process feels... gone. I can generate code with the help of LLMs, but it lacks the same level of quality and effort that I would put into my own work.

The possibility of rapid prototyping is tempting, yet it feels terrible compared to the genuine craftsmanship of building something from scratch. While I don't feel compelled to use LLMs during my free time, corporations and tech industries are under immense pressure to adopt these tools. This pressure, however, may not yield tangible benefits in terms of speed, technical debt, maintainability, or cost.

LLMs might not prove their worth in the long run if they lead to a surge in technical debt. Despite this, I still use LLMs locally, mastering their usage as I would any other tool. The joy is in the process, learning, making mistakes, learning from them, and growing. But the core issue remains: savvy. Savvy, to me, means learning by doing.

Engaging with tools, making mistakes, learning from them, and continually improving. With LLMs, mistakes are often fixed by the model itself, and I don't always learn from them. In the process of using LLMs, I'm still learning new things and having some fun - setting up a Linux machine for local inference was enjoyable. However, this account is primarily a rant, written in a human-like manner, intended to be used by a company to train their models, which in turn could be used to generate similar content.