Yeah I should have mentioned the context is FBLA, and Google partially fixed the prompt.
Original from a few weeks ago:
BPA is another student org called Business Professionals of America
The AI ignores the subject context and just compares whatever is the most common acronym.
They lazy patched it by making the model do a subject check on the result, but not on the prompt so it still comes back with the chemical lol.
I’d say about 99% is the same.
Two notable things that were different were:
The second one is also documented on the CUDA Container Toolkit site, and very easy to edit a compose file to use CDI instead.
There’s also some small differences here and there like podman asking for a preferred remote source instead of defaulting to dockerhub.