ChatGPT-4 Receives 'B' on Scott Aaronson's Quantum Information Science Final
llm-evaluationquantum-computinggpt-4benchmark
Abstraction: GPT-4 scores B on honors quantum information science exam
Key points:
- Scott Aaronson administered GPT-4 the actual 2019 final exam for his UT Austin honors quantum information science course via LaTeX source
- GPT-4 scored 69/100; student average was 74.4 (with selection bias — struggling students had dropped); equivalent to a B
- Exam max was 120 due to extra credit; highest actual student score was 108
- As a demonstration, GPT-4 was prompted to write a grade-appeal email to the Dean, citing time constraints and question style as excuses
- The experiment illustrates LLM performance on rigorous technical STEM exams approaching but not matching average human performance
Connections: GPT-4 · Openai · Scott Aaronson · Large Language Models · Quantum Computing · LLM Evaluation