The more I learn from your research studies, the less I think OpenAI can win (I’m saying this with $ on the table)

Thank you for doing both median, mode, and explaining those edge cases.

It’s funny how the outlier data you see in school – the ones you’d throw out – are in practice the hard problems that tell you what organizations are shooting to be that 99th percentile…

Because when you operate in the trillions, your errors are in the millions.

Scale always messes with my mind.

It’ll be very interesting to see what comes next on the web now that machine translations become more acceptable. Will we see model collapse or will there finally be better multilingual results and answers?

Thank you for sharing the kind of 99th percentile work as well.

*No AI was used and I should get back to sleep.

ChatGPT 5 is OpenAI’s equivalent of Google Search’s Universal Search.

You don’t sort the document types. The document types are sorted for you.

Previously we picked models with access to tools to solve for a prompt. Now, based on the prompt, the model picks the right tools for you.