I taught Claude to talk like a caveman to save my AI tokens. It became unusable — and I learned a lesson about virality.
· Business Insider
Alexander Huso
Visit asg-reflektory.pl for more information.
- Alexander Huso taught Claude to talk like a caveman to save the tokens from his Pro plan.
- While it saved some tokens, Huso said the resulting quality was poor. "I wouldn't trust it to write any serious code," he said.
- Then Huso posted it on Reddit, where he learned a lesson about going viral online.
This as-told-to essay is based on a conversation with Alexander Huso, a 31-year-old who lives in Salt Lake City. It's been edited for length and clarity.
I work for a hotel right now. I had a software testing job, and I'm a great coder, but, without a degree, it's really hard to prove that. I do it for the love of the craft.
You could really use AI for anything. When I'm looking for work, I have a ChatGPT scheduled thing. In 2026, applying for jobs is a nightmare. The best way to apply for a job is to apply right when it's posted, so you're the first one on the list. I have a ChatGPT thing scheduled every 4 hours that searches the web for obscure job boards and finds the ones posted 15 minutes ago.
I had a project recently based on incidental learning. The idea is, if you're exposed to something repeatedly without thinking about it, you'll learn it. I wrote an app to replace random words on your phone with Spanish translations, so you can learn Spanish.
I was coding for years before AI. Right when it finally clicked was like, boom, ChatGPT went viral. I'm still trying to figure out: Did I learn to code at the best time or the worst time?
ChatGPT feels unlimited sometimes; I rarely hit my limits. I feel Claude is a lot higher in quality, but the tokens you hit really fast. For things that are really important, I tend to use Claude. ChatGPT is really for anything.
I was running out of Claude tokens. I'm on the plan that's $20 a month. I was always exploring different ways to make this cheaper.
When you're talking to AI, you sometimes find yourself slipping into caveman. I used to say: "You write code." My idea was like, why don't we skip articles and different parts of speech?
My first idea was baby talk, but caveman talk wrote better, and it was funnier.
Alexander Huso asked Claude how many tokens he'd save from caveman speak.Alexander Huso
Originally, I was using it to break some bug bounties. You can ethically hack something with permission, and if you find a vulnerability, they'll pay you for it. They're going to take a lot of work, so I shouldn't expect to get one in a day. It was something I was dabbling with.
I was dropping APK files into Claude Code, and being like, "See what you can do with it." That's what I mostly used the caveman mode for: hacking Android apps.
I have not gotten a bug bounty yet. Caveman talk hurts the quality a lot. Anytime there's something important, I don't want to use it. "Me write code" is not as good a chain-of-thought reasoning as it would have been by default.
I wouldn't trust it to write any serious code.
As an afterthought, I threw it on Reddit. One of my favorite YouTubers made a video about my idea, which messed me up for a couple of days.
My GitHub is the most important thing in the world to me. If I died tomorrow, my friends and family know they should go look at my GitHub, not my Facebook or my Reddit.
I got like four GitHub followers out of the whole thing. Some kid from the Netherlands blew up and went totally viral. I wish him well. There's no such thing as stealing in open source. At the end of it, I thought: I should feel validated and flattered. That's the healthy way to look at it.
I may be a bit eccentric, and my family thinks I'm nuts. If anything, this has validated me.
My mom doesn't think I'm crazy anymore.
Read the original article on Business Insider