AI Mondays - Apple Vision Pro

AI Mondays - Apple Vision Pro

Who are we?

Ray intro worked at Apple for 12+ years and went from working at the Genius Bar to being in the studio with Jony Ives and being in the office with Tim Cook before presenting the Apple Watch.
Ronald Mannak

Rapid Fire

Ray to introduce what this is and what we are doing.
AI headlines that caught our eye. Early trends. Prototypes. Breaking News.

Trend for the week - Apple Vision Pro UI/UX Skate to where the puck is going

  • Skate to where the puck is going
  • UI/UX - too much info in app
  • Eye Fatigue
  • Re-thinking web UI
  • UI refinements with your eye

Special Guest Nisten

LAION - newer dataset, baklava model. Fully open source
Eric Topol - TED talk healthcare 30 min call - take images of retina
ChatGPT - divided into 32 layers - each part digests it a little bit then out comes one token
Models like Mistral 70 and 80 layers - LLM
Image Models - Clip encoder - OpenAI published in 2021 - output embeddings (raw tokens). Not exactly words. Built their own based on the lava method, took clip encoder based on mistral then created a new later, trained it for 80 hours on 8 A100’s. The image brain and the language brain started to understand each other.
  • Take an encoder that outputs embeddings, raw tokens
  • This thing said such and such