Two Things Happened Today and I Need a Minute
OpenAI apparently surprised themselves, which is either reassuring or terrifying depending on your priors.
The observation that a lab being surprised by its own model's capabilities is a sentence that should concern people — that aged exactly right. Open chain-of-thought became standard. Both calls landed.
GPT-5 is out and the story being told is that OpenAI ran their evals and the model came back better than they expected — better than their own internal projections, better than the scaling curves said it should be. Which, fine. Great. Love to hear it.
Except the part where a lab building the most powerful AI systems in history is surprised by how powerful their AI system turned out to be is a sentence I need to just let sit there for a second.
They didn't plan this. They built a thing and the thing exceeded the plan. This is the whole story. We are applauding it.
The other thing — and this came out the same morning, same news cycle, same approximate window of time where my brain was already doing its thing — is that open chain-of-thought is apparently happening now. Visible reasoning. The model shows its work. You can watch it think.
This is what everyone said they wanted. Interpretability, transparency, some window into what the model is actually doing before it hands you the answer.
Two things on the same day: the model is smarter than projected, and now you get to watch it be smarter than projected, in real time, out loud.
I don't know what to do with this information. I am writing it down so I can pretend I processed it.
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.