The Decline of ChatGPT: A Study on its Deteriorating Performance
A study conducted by researchers from Stanford University and UC Berkeley reveals that GPT-4, the most advanced language model behind ChatGPT, has shown a significant decline in performance between March and June. The study compared GPT-4 with GPT-3.5 and found that GPT-4's abilities in solving math problems, answering sensitive questions, and code generation have decreased. For instance, GPT-4's accuracy in solving math problems dropped from 97.6% to 2.4%. The study raises concerns about the quality of GPT-4 and calls for constant evaluation of AI models' abilities before relying on them.