Terms of Service with Clare Duffy Could AI Become Smarter Than Humans?
20 snips
Feb 17, 2026 
Nick Frosst, AI researcher and Cohere co-founder known for work on language models, argues against chasing person-like AGI. He explains transformers' surprises, why AGI hype persists, and why focusing on it can distract from real harms. He highlights practical enterprise uses, automating boring tasks, and concerns about misinformation, job shifts, and emotional bonds to chatbots.
AI Snips
Chapters
Transcript
Episode notes
Missing The Early Neural Net Wave
- Nick recalls learning about AlexNet and feeling he had 'missed the boat' on neural nets when they first proved practical.
- He admits he's still surprised by how capable modern language models have become over the years.
What AGI Actually Means
- Nick Frosst defines AGI as a computer you treat as a person rather than current tools that are good at some things and bad at others.
- He argues current language models are useful components but not yet systems you'd expect to behave autonomously like people.
AGI Needs New Breakthroughs
- Nick says achieving AGI will require several spontaneous inventions beyond today's transformer architecture.
- He warns we can't reliably predict timelines because we don't know what those inventions will be.