The Weekly Show with Jon Stewart AI: What Could Go Wrong? with Geoffrey Hinton
607 snips
Oct 9, 2025 
Geoffrey Hinton, the 'Godfather of AI' and a Professor Emeritus at the University of Toronto, joins Jon Stewart to dive into the fascinating world of artificial intelligence. They discuss how modern AI differs from traditional systems and explore the risks posed by advancements in technology. Hinton highlights concerns over AI's potential misuse, including election interference and weaponization. He emphasizes the urgent need for regulation and reflects on the uncertain future of AI as it approaches superintelligence.
AI Snips
Chapters
Books
Transcript
Episode notes
Next-Word Prediction Builds Meaning
- Language models learn by predicting the next word from context, building feature activations like human meaning.
- Hinton argues predicting next words and backpropagating errors produces internal representations that capture semantics.
Shape Models With Human Reinforcement
- Human reinforcement (labeling outputs good or bad) guides model behavior after unsupervised learning.
- Hinton says operators can shape models' outputs by reinforcing desired responses or punishing others.
Weaponized Persuasion Is An Immediate Threat
- The most urgent risks are misuse by bad actors for persuasion, disinformation, and targeted manipulation.
- Hinton ties real-world dangers to availability of detailed personal data for targeted influence.
