How I use Claude even though I hate using LLMs
The Internet is full of breathless articles about how to use AI to massively increase your productivity. These articles are all written by people who really like using AIs.
I hate using AIs, so I hate all those articles. They’re like “you can have long conversations with the AI in which you refine your thinking about the most important issues!” “whenever you’re having a feeling, you can tell an AI about it and then the AI will give you useful life advice!” and I am like “I would rather spend an hour cleaning out all the six-month-old leftovers from the back of my fridge.”
However, I have found AIs to be useful in some cases, so I thought I could do my part to counter selection effects by writing an article about how to use AIs for people who hate using AIs.
Disclaimer: I have yet to figure out any use case for Claude Code in my own life, so this article only covers the use of large language models for chat.
I STRONGLY recommend using Claude. As far as I can tell, the general consensus is that the Claudes have the best personalities of any large language model: kind, earnest, curious, and even funny. They’re just cinnamon rolls! You’re much less likely to use an LLM if you have to deal with ChatGPT’s faux-cheerful Customer Service Voice. Grok significantly underperforms other state-of-the-art models. I don’t recommend using Gemini for ethical reasons.
If you find your Claude’s behavior unpleasant, you can go into the system prompt and tell him not to do whatever you find annoying. Here is my system prompt, which I assembled based on several system prompts available online:
