Some notes on AI
I am very skeptical of AI. I also use it heavily on a near-daily basis.
Currently, I have three paid accounts running: Claude Code through my work, ChatGPT Business through overkill1, and Gemini Pro, which I pay for myself (it comes with 3TB of Google Drive).
On top of that, I have a bunch of credits loaded in OpenCode Zen whenever I want to try one of the other models (e.g. Qwen, Kimi K2), and because I use Kagi, I somehow also have AI there.
I primarily use LLMs for three things:
- Whenever I want to look stuff up where I think a simple web search might not be enough, I use Gemini (but I am also experimenting with Kagi for this). The advantage of an LLM in this scenario is that it can look at multiple sources much quicker than I ever could. But I also regularly ask Gemini if it's sure that the info is correct, and then use the sources it finds to double-check. I also use it to summarize YouTube videos.
- I use ChatGPT to look for typos in my writing. I write primarily in Obsidian, and since it's all just text files, I spawn OpenCode in the terminal and use a skill that checks for any typos, grammar issues, and syntax errors. I explicitly disallow any major rewrites or changes to voice and style. In 80% of cases, whenever I run this skill, the most it does is move a comma or fix some Markdown.
- I use Claude Code for coding things. And I use "things" deliberately, because sometimes I use it to work on Excel files, sometimes I just ask it for SQL or Python code snippets whenever I am in "data analyst" mode, and sometimes I build internal tools without actually writing many lines of code. (I also use GPT 5.4 through OpenCode for a bunch of vibe-coded apps or personal changes to open-source projects.) I know my way around Python (primarily through pandas), and I can read most code, but I am simply not a good developer, and never bothered improving that skill. And since it's all internal tools either for me or for the company, I don't particularly care if the code is all spaghetti.
But there are two things which I will never use AI for: writing, and therapy/thinking.
In fact, this is the usage of AI that drives me the most insane whenever I observe it in other people. (Or read about it on their blogs.)
Hacker News is filled with blog posts in which the author complains that their skills and critical thinking have been atrophying since they started using AI.
Recently, I read this (not linking to it, because I don't want to criticize the author directly):
This article isn't written with AI but I just caught myself about to copy and paste it into Claude to see what it thinks because I'm worried that it doesn't make sense or it reads funny or there's something missing. That's the self-doubt that it's feeding on and what I need to fight back.
I don't get it. By now, enough people should have heard that LLMs do not actually improve your thinking, but are instead preprogrammed to hype you up. Which, funnily enough, we see most in CEOs. LLMs are the most personal, most effective "yes men" you can think of. These motherfuckers are all suffering from some sort of psychosis.
The other part that I struggle to understand is why people think they need to write blog posts (especially on their personal blogs) with AI.
First of all, a personal blog is a creative outlet. The goal of that outlet is the act of creation itself. It is the reason why I removed every kind of analytics from my blog the moment there was no monetary incentive anymore.
What do I care about how many people read this website? I primarily write it for me and for the need to express myself. If some people read it, cool. If no one does, I wouldn't even know in the first place.
Writing personal blog posts with AI is like buying a keyboard with a self-playing function. You wouldn't hit play and then pretend you just made music while the keyboard plays itself?
It also doesn't help that all AI writing reads the same. It doesn't matter whether it's social media posts, blog posts, YouTube scripts, or emails. AI-written text has a certain "style" that I find tremendously off-putting.
As soon as I stumble upon such writing, I immediately lose trust in the arguments made by the "author." It's a subconscious process. It doesn't matter if the arguments are sound, I simply cannot trust it anymore.
It's like smelling a fart. Doesn't matter how nice the person is. You cannot unsmell the fart.2
Finally, and I might be completely naive here, I do not see AI completely replacing human labor anytime soon.
So far, no matter what the thing is, you always need a human in the loop to actually ship what AI did. It might get 80% of the job done, but the remainder is where stuff gets tricky.
Also, a lot of the replacement hype seems limited to the coding world. And a lot of it is due to the human fart that is Dario. Turns out that if you have a financial interest in making people believe your shit is better than it is, you keep pretending it is better than it is. That's like Lucky Strike saying that a cigarette is better for your health than a sweet.
Now, of course, that doesn't help all the thousands of people who were laid off. Although I wonder how much of this is truly because of AI, or because Silicon Valley just jumps on any excuse to save money and fuck with their "human resources". AI just happens to be the excuse of the week.
But I do believe that juniors will have a harder time getting started since the jobs juniors usually do are now done by LLMs. I don't know how this will change.
Anyway, this is an excellent interview that explains why I am not as pessimistic as some of my peers.
(Or if you prefer a written version.)
Finally, as for the thinking part, I have written enough in my post titled How to think. I still stand by it.
But if I have to add one thing, it's that people often forget that LLMs are basically a mix of pirated books, TED Talks, Reddit comments, Twitter posts, and porn.
Tell me how that makes for great thinking?
Or as Mo put it:
We thought AI was gonna make everyone a genius. What it's actually doing is taking the smartest people in the room and moving them toward the middle. It makes average sound like genius.