Short videos on AI/LLM Limitations and “Hacks” by Alberta Tech
-
Dan MacIsaac [1]
-
[1] Column Editor Physics Department, SUNY-Buffalo State University, Buffalo
-
- Localización: The Physics Teacher, ISSN 0031-921X, Vol. 63, Nº. 1, 2025, págs. 77-77
- Idioma: inglés
- Texto completo no disponible (Saber más ...)
-
Resumen
TikTok and YouTube author and tech/AI comedian Alberta Tech has a series of amusing and insightful short presentations on limitations and hacks that can be applied to AI and LLMs. Alberta introduced me to the now well-publicized “How many R’s are in Strawberry?” challenge that LLMs struggle with. For interesting reading, google that question, then scroll down. Both Grant Sanderson and TechCrunch explain what’s happening quite well. I’m gradually coming to the conclusion that I have to help my students learn how to very carefully and skeptically operate LLMs to produce useful physics. It’s really clear that LLMs produce word salad that makes customers go away happy, and their physics output is often wrong, usually incomplete, wholly untrustworthy, and sometimes gripped by hallucinations. Sabine Hossenfelder describes the problem as one of the language of words vs. mathematics and logic
