ChatGPT, the leading generative AI offering, still experiences a lot of hallucinations. An AI hallucination refers to the occasions when AI generates false or misleading information that appears reasonable but is not based on facts. This past week, I have been intentionally conversing with the generative AI tool to experience its strengths and weaknesses.
During our exchanges, I have provided it with information to update its memory with my favorite NBA players, soccer teams, books, and music personalities, aiming to shape and optimize its output for me. I also used it to schedule my daily to-do lists and set monthly reminders for documentary suggestions.
While it is an excellent tool for many of these tasks, it is alarming how many times it hallucinated and the type of information it misreported. For example, earlier in the week, the biggest news in the NBA world was the trade of Luka Doncic to the Lakers from the Mavericks. This was the talk of the town for basketball fans. While conversing with ChatGPT, it failed to recognize that the trade happened – until I asked it again the same.
I also asked it to give me score updates for the ongoing soccer match. It failed to recognize that the game was ongoing and later on gave me misleading score updates. Because of this, I’m worried about whether I will receive my set monthly documentary suggestions.
These instances led me to question the implications of the generative AI tool in the education sector. This can lead to serious inaccuracies for students who might be harnessing the tool for research purposes as an alternative to Google Scholar or as the primary search engine. Consequently, it demonstrates a) the vitality of human agency in AI integration, b) the need to foster critical thinking and deeper inquiry while using AI tools in education, and c) the importance of continuously improving AI models to enhance their accuracy and reliability in knowledge-based applications.
The tool’s tendency to provide accurate output in the second attempt is great, but how would users know when to re-query it? Especially when dealing with often complex mathematical or scientific subjects.
1 comment