ChatGPT-4.5 Did Something I Didn't Expect | Medium
I had my first experience with OpenAI’s new model, ChatGPT-4.5 that dropped today. I expected a better, more refined model. What I observed, seemed like so much more.
It all started with my simple prompt, “Create a compelling X thread and poll based on my latest YouTube video: ‘90% to AGI — Why Did the Counter Just Tick Up?!’” ChatGPT delivered a solid post — well-structured, engaging. But then, I saw something strange. It referenced a platform called “Instagrain” — an AI-generated social network. That caught me off guard. I’ve never mentioned Instagrain, never heard of it. It’s not in my video, nor does it exist.
It offered to research whether or not Instagrain existed. But even after I told ChatGPT-4.5 that it didn’t need to investigate Instagrain, it did anyway. Not in an irrelevant or stubborn way — but it seemed like it wanted to understand what went wrong. ChatGPT traced the error, found the root cause (a mix of typos, misprints, and obscure references online), and broke it all down for me — step by step.
This is the part that really got me. Most AI models would have corrected the mistake and moved on. They wouldn’t have researched anything if I specifically said not to, and that would’ve been the end of it. But this model didn’t just fix the mistake — it wanted me to understand what happened, too.
For an AI that supposedly doesn’t reason, this seemed like reasoning to me. It showed a self-reflective quality, making sense of its own process in a way that I haven’t seen with other LLMs. Sam Altman was right — ChatGPT 4.5 was different. Its responses were intentional and nuanced.
Further Insights
If you’re curious, I’ve posted the full transcript of my conversation with ChatGPT-4.5, so you can see exactly what I mean. And, if you want to dive deeper into AI’s accelerating march toward AGI, check out my latest video:
