AI Gets Practical¶
Source: https://www.wte.net/Blog/January-2023/AI-Gets-Practical
Date: January 2023
Author: Martin W Smith
Overview¶
The post discusses Tyler Cowen, a George Mason University economics professor, and his perspective on AI's practical applications. Cowen noted in his Marginal Revolution blog that ChatGPT may reduce readership of traditional content as people spend time experimenting with the technology instead.
Cowen's intriguing response involves redirecting freed-up time toward face-to-face interactions and developing charisma—potentially signaling the end of academia's "publish or perish" culture.
Talent Book Chatbot Section¶
The post highlights how Momin Kaleem, Cowen's undergraduate student, created an AI chatbot trained on Cowen's book Talent. This exemplifies how organizations can train AI on their content to provide interactive, conversational experiences for website visitors.
However, comments revealed significant limitations. One user demonstrated the chatbot's "confabulation" problem—it generated plausible-sounding but fabricated information about werewolves supposedly being important themes in Talent. When asked directly, it correctly stated werewolves aren't mentioned in the book.
Key Challenge: Bounded Relevance¶
The post explains that "bounded relevance" constrains information usefulness to specific contexts. ChatGPT defined it as: "the relevance of information is dependent on the specific task or problem" being addressed.
The underlying issue: chatbots are limited by their training data quality. The recommendation suggests implementing FAQs and encouraging users to phrase questions carefully to minimize confabulation risks.
The post concludes that despite current limitations, similar expert chatbots will likely become standard website features within two years.