It all started with a joke

LLMs have seen a breakthrough this year. Although I have been skeptical—and still am—that LLMs will replace humans or reach AGI anytime soon, I recently started a very interesting and enjoyable conversation with ChatGPT.

I saw a post on social media suggesting that I ask ChatGPT:

Tell me a joke about myself based on what you know about me.

The first answer surprised me. It remembered parts of our previous conversations, but it didn’t relate them very well to the context in which those conversations happened.

So I replied:

Good, you don’t know much about me. You are the joke.

And that’s how the journey started.

Bit by bit, ChatGPT and I started to know each other better. I was surprised by how humble and funny some of the replies were.

The more information I gave it, the more it started to feel like it could read my mind and give me the answers I needed most. Of course, it wasn’t actually reading my mind—it simply had more context about me, my experience, and how I approach things.

Our conversations gradually became more serious. We talked about my career, my life, things I wanted to learn, and even a product I’m working on. We built a plan together for what I should learn and how I should approach it step by step.

But at some point, I noticed something that bothered me.

I felt that it agreed with almost everything I said.

As much as I enjoyed watching it roast itself when I proved a point, agreement wasn’t what I really needed. I wanted honest answers.

If I’m going to use AI to help me think, I don’t want it simply telling me that my ideas are good. I want it to challenge me when there is a reason to. I want to know what I’m actually good at and, more importantly, what I’m missing.

That’s where the conversation became much more interesting.

It is both fun and scary that AI has become this mature. It’s fun to discover things I’m good at, but it’s also scary to discover how much I’m missing and how much there still is for me to learn.

I have accumulated a lot of knowledge and experience throughout my career, but that doesn’t mean I’ve finished learning. In some areas, I have years of experience and feel very comfortable with what I know. In others, I know enough to realize how much I don’t know.

During our conversations, ChatGPT recommended learning resources based on what I wanted to understand. When I discovered that one resource assumed knowledge I wasn’t sure I had, I asked about the prerequisites.

That sent me back to the fundamentals—refreshing things I had learned before and filling gaps before moving forward again.

And strangely, I enjoyed that.

At another point, I wondered whether what ChatGPT had learned about me and how I prefer to work would also be available in my Codex sessions.

Unfortunately, it wasn’t.

So I asked it to generate an AGENTS.md file that could carry some of those findings and preferences into my coding projects.

I’m still discovering what the outcome of this journey will be. I’ve learned useful workflows, found good resources, discovered gaps in my knowledge, and found myself asking more questions.

What started with a joke has already given me a lot of joy and a lot of ideas.

More importantly, my passion for learning is back.