site banner

Friday Fun Thread for September 27, 2024

Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.

2
Jump in the discussion.

No email address required.

A completion model called gpt4t-lu-test just appeared in OpenAI's API Playground (and presumably the OpenAI API generally), and it's not documented anywhere (including in Pricing), and ~nobody seems to have noticed yet. It doesn't seem to even be instruction-tuned, and will gladly complete a meth recipe etc.

As far as I know, OpenAI has never allowed any public access before to a GPT-4 base model, but that seems like what this is.

What is this, and what does the lu there stand for?


update: gpt4t-lu-test is no longer available as a completion model, but now as a chat model, that seems at very shallow first glance to just be a normal GPT-4 chat model (guardrails and all). I guess I caught a brief misconfiguration.


update: Now gone entirely.

Employee claims misconfig, and nothing special. I buy that because it seemed unlikely they would be doing something interesting off of Turbo at this point.