4 Comments

Hi Tom, thank you for sharing. Large Language Models are ill suited to tasks where accuracy is important. Using them as tutors or to give feedback are misuses of the technology. As a veteran of ten years as a Special Educaton teacher, I shudder to think a teacher uses an LLM to "provide feedback for her "high needs" students."

Expand full comment

Thanks, Tom. Sorry I missed your comment earlier.

So, there are plenty of valuable uses of AI as a tutor. I have spoken to science, world language, computer science, ELA, and elementary teachers who would tell you that. GTP-4o is a great Spanish tutor for my goals. But that doesn't mean an AI tutor is right for every student in every situation. LLMs are improving, but human guidance remains critical.

Expand full comment

We're always told to verify LLM-generatred text. As if teachers have time for that. How does that work when a student reads text generated by an LLM tutor? Are students supposed to verify every claim in realtime during a tutoring session?

Expand full comment

Please read the works of Gebru, Bender, Mitchell, Hanna, et. al. They have researched LLMs for years. They are not saying that LLMs are appropriate as tutors. They are saying just the opposite. https://www.criticalinkling.com/p/teachers-follow-these-experts-to-learn-ai

Expand full comment