Better Offline The Academics That Think ChatGPT Is BS
48 snips
Jul 19, 2024 
Academics from the University of Glasgow discuss their paper classifying ChatGPT's outputs as 'BS'. They explore AI communication nuances, language learning, ChatGPT's limitations, and the challenges of using AI in academia. The conversation dives into the shortcomings of AI models and the dangers of overhyping them, emphasizing the importance of human intelligence in problem-solving.
AI Snips
Chapters
Transcript
Episode notes
ChatGPT and Bullshit
- ChatGPT exhibits soft bullshit, speaking without regard for truth.
- Hard bullshit requires deceptive intent, potentially applicable to designers or prompters.
ChatGPT's Lack of World Representation
- ChatGPT doesn't represent the world; it statistically models human language.
- It lacks the perception and tracking of external world entities essential for genuine representation.
Expletive Infixation Example
- James Humphries highlights ChatGPT's inability to grasp expletive infixation.
- He uses the example of inserting "fucking" into a sentence, a task easy for humans but difficult for ChatGPT.