🤖Murati Previews 200ms Real-Time 'Interaction Models'
TL;DR
Mira Murati used her first interview in 18 months to show Thinking Machines Lab's bet. Its models process audio, text, and video in 200-millisecond intervals instead of turn-based prompts.
Mira Murati used her first interview in 18 months to show Thinking Machines Lab's bet. Its models process audio, text, and video in 200-millisecond intervals instead of turn-based prompts. She called it a first step, with no release date.
Key Points
Thinking Machines has shipped one product so far, Tinker, an API for fine-tuning open models
Interaction models target real-time streams in 200ms chunks to catch interruptions and pauses
Murati warned against concentrating AI decisions in too few hands across the industry
She declined to give a ship date or call the work finished
Why It Matters
If continuous real-time interfaces work, they could break the turn-based prompt model that every chatbot is built on today.
Quick Facts
Frequently Asked Questions
Why does this matter?
If continuous real-time interfaces work, they could break the turn-based prompt model that every chatbot is built on today.
What happened?
Mira Murati used her first interview in 18 months to show Thinking Machines Lab's bet. Its models process audio, text, and video in 200-millisecond intervals instead of turn-based prompts.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.