Can You Guess Which Song Snippets Are Written by AI?
summary
Shows GANs trained on image and lyric corpora to generate faces, dogs, paintings and song snippets and runs a two‑question audience quiz where the line "I'm in love with the shape of you we push and pull like a magnet too" turns out to be real while the other lyric is AI. Explains generator vs discriminator, loss/backprop, common glitches (weird backgrounds, warped teeth, asymmetrical glasses, floating accessories), points to thispersondoesnotexist.com, uses GANs, and warns that 95% accuracy needs huge datasets and long Colab training.
AI-written summary · about
chapters (14)
- 0:00Introduction and welcome
- 1:09Interactive quiz begins
- 5:33Explaining GANs theory
- 12:19Identifying AI-generated faces
- 16:48More AI-generated face examples
- 21:55Analyzing AI-written lyrics
- 24:21Other AI-generated content
- 25:47How GANs work recap
- 27:12Resources and code discussion
- 27:32Q&A on data sets and accuracy
- 28:38Discriminator misclassification
- 29:43Deepfakes and AI risks
- 31:11Real-world uses of GANs
- 32:13Research on detecting AI content
related talks
28mLyrics Generator + Results: AI Usecases for climate change
03mBuilding artificial people: Endless possibilities and the dark side
35mGenerative AI in the real world
59mComputational Creativity
44mWith AI, Does Music Have the Blues?
53mGenerative AI: Overview and use cases
28mGPT or how I learned to stop worrying and love the generative AI
45mGenerative AI for Image and Video Production