Artificial Intelligence. Our journey towards singularity
summary
A practical survey of what current systems can and cannot do: a car classifier called Cyclops reaches 98% accuracy on fine-grained make/model/trim after training with 1 million images using TensorFlow and Inception v4 and running on 10 GPUs for three weeks on AWS. Covers transfer learning, VAEs and GANs for face generation, morphing and super‑resolution, and argues we lack orchestration, cross‑domain transfer, reasoning and true self‑learning for AGI.
AI-written summary · about
chapters (29)
- 0:00Intro
- 0:26A LITTLE BIT ABOUT MYSELF
- 1:07GENERATING BABY NAMES
- 4:32OUTLINE
- 5:19HOW WE SOLVE PROBLEM TRADITIONALLY
- 6:24WHATS THE PROBLEM WITH THIS?
- 7:33SOLVING PROBLEM THE AI WAY
- 7:48STEPS
- 9:22ARTIFICIAL INTELLIGENCE
- 9:46MACHINE LEARNING
- 10:17DEEP LEARNING
- 14:05Convolution Neural Network
- 14:53Inception V3
- 17:26Transfer Leaming
- 17:59TRANSFER LEARNING
- 18:55DISTRIBUTED TRAINING
- 19:17TRAINING CYCLOPS
- 20:57IMAGE GENERATION
- 21:32AUTOENCODER
- 25:22BIG LUGGAGE
- 26:10MEMORY GAMES
- 27:24LATENT SPACE DISTRIBUTION
- 29:03MORPHING
- 29:18LATENT VECTORS
- 31:33ADDING GLASSES
- 31:54REMOVING GLASSES
- 36:58GAN PRO AND CON
- 38:34SINGULARITY
- 43:49UNDERSTANDING IS NOT A REASONING
related talks
13mArtificial General Intelligence: Why aren't we there yet?
45mThe Curious Incident of Developing Artificial General Intelligence
47mThere is no such thing as Artificial Intelligence
18mTrinity of Artificial Intelligence
36mArtificial Intelligence: The Future of Software
47mSponsored Session: Beyond the Buzz: Navigating AI's Myths and Realities
19mAI: The current state of play
01h 14mLatest trends in AI