Applied AI Pod

Game up: AI models, GPT, and Cyberpunk 2077, with former AI teacher, Virgil Ilian, E22

Episode Summary

AI today is a catchall term. I’ve invited Virgil to catch a bit on some of the hottest 2021 trends, games and moves. Virgil Ilian, PhD - Senior Research & Strategy Consultant (Artificial Intelligence, Blockchain) is an educator and techie having fun at the limit of art and technology. He is a former AI teacher that currently helps the industry to bridge AI and Blockchain research and work in their products and initiatives. He also helps build an ecosystem in his role as Program Manager at BeAI, Europe's AI Pre-Accelerator.

Episode Notes

• 2:00 - Hottest AI trends for 2021
• 5:35 - Open source for AI - paradigm shift
• 11:30 - AI model supremacy
• 21:10 - Authorship rights when AI contributes
• 31:00 - GPT encapsulating knowledge?
• 34:00 - Human consciousness replicable as computation
• 41:50 - Are we in a matrix?
• 42:30 - Cyberpunk 2077
• 50:50 - Can AI create emotion the way we cannot tell it is AI?

Conversation references:

• Book: "You look like a thing and I love you" - Janelle Shane
• Book: "Shadows of the Mind", Roger Penrose
• Chinese Room argument
• Manhattan project
• Art Breeder project
• The Origin of Circuits - re FPGA topic

Host's notes:

• Gartner Top Strategic Technology Trends for 2021 
• Jukebox - music-making tool by OpenAI. While the achievement is significant from a technological perspective, the results are unlikely to threaten the livelihoods of human musicians.
• DALL·E generates images in response to written inputs, and (whose name honours both Salvador Dalí and Pixar’s WALL·E) is a decoder-only transformer model. From Andrew Ng's 'The Batch' newsletter: OpenAI trained it on images with text captions taken from the internet. Given a sequence of tokens that represent a text and/or image, it predicts the next token. Then it predicts the next token given its previous prediction and all previous tokens. This allows DALL·E to generate images from a wide range of text prompts and to generate fanciful images that aren’t represented in its training data, such as “an armchair in the shape of an avocado.” WHY it matters? As Ilya Sutskever puts it ‘combining language and vision techniques could overcome computer vision’s need for large, well labeled datasets’.