Please note: This seminar will take place in DC 2564.
Daniel M. Berry, Professor
Cheriton School of Computer Science
Problem: This talk tries to explain why so many perceive large-language models (LLMs), such as ChatGPT, as intelligent and creative.
Observations: An LLM generates convincingly cogent bullshit (BS) in the Frankfurtian sense. Humans instinctively perceive the ability to BS convincingly as an honest sign of intelligence. Each claim is supported by empirical evidence.
Conclusion: Therefore, humans instinctively perceive an LLM as intelligent.
Observations: A creative idea can be an exception to the norm that is perceived by humans to be a good idea after all. An LLM’s BS contains mistakes, some of which are perceived as creative.
Conclusion: Therefore, humans perceive an LLM as creative.
Observations: Fundamentally, an LLM is a large model of a language, no less and no more.
Conclusion: Therefore, it is no surprise that asking an LLM to do any task but a strictly linguistic task yields BS!