2025.02.15 Citrus Mini-Survey

I bought 5 varieties of Citrus from my local Maruichi and then asked each of my friends to rate the citrus and then write the rating on the back of the packages.

I then gave GPT the images and asked it to give a correlation matrix of people's preferences. It did surprisingly well!

Here are some mistakes that it made:

It didn't pick up BC's ratings, and upon being prompted to add BC data, it got 1/5 correct. (This is basically about as good as random chance, given the rating window).
It got 4 (out of 35) of the other ratings incorrect - (JX, Ikoyan), (AS, Amanatsu), (JJ, Kawachi Bankan), (AS, Ikoyan)

It's a picture. I'm too lazy hehe.

Lmfao, basically Eric is the only person that enjoyed Amanatsu and his correlations show it.

KL and MY rated simultaneously and probably had some info leakage.