
Artificial
New Leader in the GPQA-Dumb Model Benchmark
Introducing Bongochat, the current leader globally in the GPQA-Dumb category, where the lower the score the higher it’s weighted. Repo/open-weights: https://github.com/ninjahawk/bongochat When asked to solve the unified field theory, it repeats the word theory back to you 50 times. It doesn’t remember anything. When solving the Math-500, it didn’t realize it was supposed to answer the questions so they were basically all blank, besides that it always did A. For coding it got 0/500. ...
Read the full discussion on Reddit
This article was aggregated from Reddit. Click to join the conversation.
View on Reddit