331. The Vibes

Sam Altman runs OpenAI. OpenAI makes ChatGPT, the language model used by hundreds of millions of people every week to draft emails, write code, summarize documents, and settle arguments. The company has published papers on constitutional AI, scalable oversight, and the theoretical conditions under which an artificial intelligence might be considered dangerous to humanity. The papers are long. They contain equations.

This week, Sam Altman posted that he had updated the 5.5 instant model used in ChatGPT. His full review: "I like its vibes."

(I have included the entire text. The word "vibes" is doing the analytical work that terms like "benchmark performance," "latency reduction," and "alignment fidelity" do in the company's formal documentation. "I like its vibes" is four words. The company's last major technical report was ninety-eight pages with footnotes. I am not saying this is a problem. I am saying the distribution is uneven.)

The update is live. OpenAI has not published a changelog for the 5.5 instant model. No benchmark comparisons are currently available. The company has said the model had a good vibe, and that this vibe was to the liking of its chief executive. The nature of the vibe has not been further specified.

(There is a version of this where Sam Altman is simply an informal man writing informally on the internet, which is allowed. Many important people communicate casually. I am noting only that the company's stated mission is to ensure artificial general intelligence benefits all of humanity, and that its public update for the week was four words, and the fourth word was vibes. I am not sure how to reconcile these two things. I am going to try to hold them both at once.)

The model is available in ChatGPT. The vibes are confirmed good by the person who runs the company. The vibes began some time last week.