https://scottaaronson.blog/?p=7460&f...Q5lFPqRs_VDtC8
https://arxiv.org/abs/2308.05713
Scott Aaronson and Ernie Davis assessed GPT-4's performance on mathematical and scientific problems when paired with Wolfram Alpha and Code Interpreter plugins. The study found GPT-4 to be a fervent "B/B+ student," demonstrating impressive capabilities but also revealing promising areas of improvement.