Theories of Everything with Curt Jaimungal Epistemology of Chatbots | Steven Gubka
8 snips
Jul 2, 2024 
Steven Gubka, a postdoctoral associate specializing in the ethics of technology at Rice University, dives into compelling discussions about the intersection of AI and human emotion. He tackles the misconceptions surrounding language models, emphasizing their limitations and the dangers of anthropomorphism. Gubka explores the importance of building trust in domain-specific AI, especially in fields like medicine. He also raises thought-provoking questions about emotional connections to chatbots and the implications for real human relationships.
AI Snips
Chapters
Transcript
Episode notes
Chatbot Anthropomorphism
- People tend to anthropomorphize chatbots, viewing their mistakes as hallucinations or confabulations.
- This reductive anthropomorphism hinders critical thinking about LLMs and their limitations.
LLMs as Tools
- Treat large language models as knowledge tools, not epistemic agents with beliefs.
- Recognize that chatbots don't hold beliefs and shouldn't be viewed as testifiers.
Unpredictable Improvements
- While additional data and training can improve LLMs, they don't guarantee increased reliability.
- Unexpected biases and performance degradation can emerge alongside improvements.