Rethink Rethink... the promise of AI
Jan 29, 2026
Tracy Dennis-Tiwary, CUNY professor on mental health, warns chatbots are psychological first aid not therapy. Ethan Mollick, Wharton professor, explores practical AI use in work and education. Rosalind Picard, MIT affective computing lead, discusses medical and emotional AI like EpiMonitor. Mike Wooldridge, Oxford AI theorist, explains transformers and LLM mechanics. They debate risks, uses, and where AI might realistically head.
AI Snips
Chapters
Books
Transcript
Episode notes
LLMs Are Supercharged Next-Word Predictors
- Large language models are massively scaled next-word predictors built with the transformer architecture.
- They learn from vast digital data using thousands of supercomputers over months to predict likely continuations.
Fluency Isn’t Understanding
- LLM outputs are probabilistic mimics of their training data and can seem human without understanding content.
- That appearance of understanding can mislead users into overtrusting the models' responses.
Don't Rely On AI For Crisis Help
- Treat AI responses cautiously because the algorithm doesn't understand meaning or intent.
- Avoid relying on chatbots for crisis or severe therapeutic needs without human oversight.
