You don’t know what’s the real impact of something before you ditch it.
That’s what I did with AI chatbots: they are great, but recently I felt the need to go on an “AI diet” for a while.
I’m not an LLM expert, but as anyone using it for pretty serious stuff (work, coding, personal life), I started to perceive the dangers that come with the undeniable productivity gains LLM tools provide: lack of reliability, increased dependence on it to perform even some basic things, and the psychological effects linked to the constant release of dopamine they induce.
I am currently on a little sabbatical, so I have no need to be the most productive guy in the room. So, being aware of the above, I decided to step back from it to gain some clarity of mind and really understand the concrete effects it has on my brain.
Here’s what I noticed :
In programming: when I started this experiment, I was in the middle of a personal project and was assisted by a coding agent (OpenAI’s Codex).
When I jumped back on this project without this help, I felt frustrated because I couldn’t deliver new features as quickly as before. As if I had suddenly stopped eating sugar, I felt an urge to reconnect and finish my tasks.
The other thing I noticed is that my coding skills had started to regress in areas that I thought I was still mastering. Even for personal projects, I always review AI code and make sure I understand it.
But when it was time to implement new features (I had to change the state of a React component after clicking on it, something pretty basic), it took me a while to actually remember what I had to do.
I had fallen into a classic illusion of learning: thinking that reading or reviewing something means you’ve learned it.
But then, over time, I started to regain confidence and even pleasure in coding, despite (or maybe because of?) my relative slowness in execution compared to a coding agent. Learning and experimenting with new tools seemed enjoyable again, whereas I had lately started to consider tool choice a secondary question.
In personal life, I didn’t realize this point before I shamelessly relapsed and opened ChatGPT one day for a personal matter. It was 3 days after I started this challenge, and before that, I started to feel a little more peaceful about planning my things on paper, far away from the computer and its temptations.
I wanted to brainstorm on a private issue, and I decided to cheat “just to get some ideas”. Quickly, after a few back-and-forths, I started to realize that I was becoming more nervous, expecting ChatGPT to have already solved my problem (which couldn’t happen because it was an open question; there was nothing to “solve” per se).
When I realized what was happening, I turned the laptop off and decided to go for a walk instead to think about my issue. Turns out it was much more effective.
In summary, here’s what this little experiment taught me :
- It goes without saying that AI can give tremendous improvements in productivity
- But little by little, you’re going to regress in the skills that you ask AI to do for you, even if you don’t feel it. Is it bad? Not always, sometimes it might even free your brain for other things. But make sure you’re really aware of the skills you’re giving up.
- AI chatbots are also incredibly addictive in ways that you probably don’t even realize for yourself
Nowadays, it’s hard to do without AI, especially in the workplace. In programming, for example, if you use it wisely (not pushing unreviewed code to production; using it to understand a codebase, etc.), you can expect incredible speed gains without compromising on quality.
But you don’t always need to be productive. Many activities (like learning or creating) require you to go slow. So if you want to use it to improve yourself and not to regress, do yourself a favor and dedicate some time to not using it.