Deepseek-r1 on my laptop w/ GeForce RTX 3060

140 tokens per second is not too bad.

I asked the 7b model to solve the 100-door problem in Python. It came up with a pretty good solution (not as cool as in BASIC, but it’s an AI, so…)

Then I asked it to convert to Commodore BASIC. It said, “I don’t know much about Commodore BASIC, so I’ll have to think about this,” and then came up with this (pretty impressive):

I needed to show it the correct syntax for a loop and that SQR and INT were three letters (it spelled them out), but it got very close on the first try with limited data on this subject.