Language & AI, Part 1: The Mirror

Recently, I joined the polygamous AI club and exported my chat conversation history from ChatGPT to Claude.

My goal was to get fresh and critical insights on how I interact with AI. Spoiler: it was a wake-up call.

To be honest, I was hooked on GPT due to laziness. I told myself: 'GPT knows me so well, who has the time and energy to educate a new LLM from scratch?' But something didn't feel right. My prompting had become mechanical and my thinking narrow.

The data proved it: I asked GPT to 'please rephrase' my own thoughts 49 times. I knew what I wanted to say. I just couldn't find the words. I was outsourcing my own clarity.

In human interaction, we have tone, body language, facial expression, history and countless other signals beyond words. But with LLMs we are naked. We have only our words. The words we choose and the context we provide are even more crucial.

So I fed Claude 438 conversations: 6,824 messages, spanning Feb 2023 to Feb 2026. Almost exactly 3 years of ChatGPT history.

I didn't know what I expected to get, but the results left me intrigued enough to go deeper. What I found wasn't about AI at all. It was about me.