Harvard Study Reveals ChatGPT's Struggles Against Students in Cognitive Tasks
A study conducted by researchers at Harvard University has revealed that ChatGPT, a language model developed by OpenAI, significantly underperformed compared to doctoral students in a series of cognitive tasks. The study involved students from Harvard's Principles of Molecular Biology course and aimed to assess the AI's ability to perform tasks at various cognitive levels. While the researchers hypothesized that ChatGPT would perform similarly to students on lower cognitive levels, the AI struggled particularly with 'apply' level tasks, which involve identifying and rationalizing experimental controls. The study found that students outperformed ChatGPT in 'remember', 'understand', 'apply', and 'analyze' questions, with the AI earning a 66 percent average compared to the students' 87 percent. The research highlighted the AI's difficulty in multi-step, compositional thinking, despite improvements in newer models. 
