Learning From Machines
summary
Shows noisy DeepDream visuals produced by DeepDream gradient ascent through Inception layers, explains convolutional filters and the 120 million photoreceptors vs 10x fewer ganglion bottleneck and a ~10 megabit optic nerve, demos GAN-generated faces and their quirks (extra eyes, malformed cows), connects adversarial examples to optical illusions, proposes recurrence/capsule ideas, and claims LSD acts on 5-HT2A binding ~221 minutes to amplify feedback — useful if wanting concrete methods, numbers, and experiments.
AI-written summary · about
chapters (12)
- 0:00Intro
- 1:00Welcome
- 2:12Hallucinations
- 12:15Serotonin
- 15:58Visual cortex
- 19:41Optical illusions
- 21:00Fake celebrities
- 23:42Fallout Cow
- 26:01Grid Cells
- 27:42Efference
- 29:44The Cerebellar
- 32:16Conclusion
