Shypd.aiThe largest open index of AI toolsThis article reports on a study where college professors administered a science exam to ChatGPT, revealing its limitations in accuracy and consistency. The AI tool received a 'low D' grade, highlighting gaps in its understanding of complex scientific concepts.
Trending
About
This article details a study conducted by college professors who gave ChatGPT a science exam, resulting in a 'low D' grade. The research, led by Mesut Cicek at Washington State University, found that ChatGPT struggles with accuracy and consistency, often providing different answers to the same question and even flipping between 'true' and 'false' on identical claims. Errors were particularly pronounced with unsupported hypotheses, where the model showed a bias towards agreement. While initial overall performance appeared solid, effective accuracy dropped significantly when random guessing was factored out. The study emphasizes that ChatGPT, as a large language model, predicts likely next words rather than checking facts, leading to 'hallucinations' and making it a risky tool for critical decisions. The authors advise treating AI as a drafting partner, not an unsupervised decision-maker, and recommend verifying its outputs.
Capabilities
Pricing & Plans
unknown
Free
FAQs
Trending
.jpg&w=3840&q=75){kind=small}
