AI Snips
Chapters
Transcript
Episode notes
Voice Cloning Scams Already Happening
- Gary recounts voice-cloning scams where callers impersonate relatives to extort money.
- He notes only short audio samples are needed and victims, like elders, already fall for it.
Fluent But Unreliable Models
- Large language models statistically assemble words, they don't understand facts.
- That leads to confident but incorrect outputs: "frequently wrong, never in doubt."
Avoid Industry-Written Rules
- Prevent regulatory capture by keeping companies from writing the rules alone.
- Demand real transparency about training data and model internals to address bias and copyright.