It's All About AI
summary
A demo-heavy survey shows practical builds: a VGG19 classifier and DeepDream visualizations, a real‑time neural style transfer applied to live video, and real‑time deepfakes assembled using MediaPipe 368‑point face mesh plus autoencoders — they run live but break on side angles and harsh lighting. A “shitposting” system fine‑tunes GPT‑2 on scraped Twitter/Reddit and GPT‑3 answers questions, generating plausible comments. Edge deployments include body segmentation on MobileNet for GDPR‑conformant cameras and an in‑house Confluence translator running open‑source models in AWS.
AI-written summary · about
chapters (32)
- 0:00Intro
- 4:29What is innovation hacking?
- 7:27Agenda
- 7:46AI
- 12:32Computer vision in a nutshell
- 15:52Demo
- 16:27Computer vision continued
- 24:30Demo
- 25:28Computer vision continued
- 26:39Demo
- 27:40Neural realtime style transfer
- 31:40Demo
- 33:46Realtime deepfakes
- 34:50Demo
- 35:21Realtime deepfakes continued
- 35:40Demo
- 36:29Realtime deepfakes continued
- 36:48Demo
- 37:34Realtime deepfakes continued
- 39:52Demo
- 41:16NLP
- 43:25Demo
- 44:29NLP continued
- 45:17Shitposting AI
- 53:00Demo
- 55:06Shitposting AI continued
- 55:33Demo
- 57:12Conclusion
- 57:50Demo
- 59:03Conclusion continued
- 1:01:08One last thing...
- 1:02:34Outro
related talks
17mAI/ML demo derby
51mBeyond the AI Hype: What's Real, What's Next
59mKeynote: The past, present, and future of AI for application developers
43mBeyond the AI Hype: What's Real, What's Next
58mPushing Deepfakes to the Limit-Fake video calls with AI
51mOpenAI and the AI that may overtake humans
56mDeep Fakes 2.0 - How Neural Networks are Changing our World
58mKeynote: The past, present, and future of AI for application developers