"Steal This Talk"
summary
Builds Talkie, a recurrent neural network that trains on hundreds of Ruby talks using textgenrnn on Google Colab GPUs, brings training loss down to about 2.9, and generates plausible talk titles and a full abstract. Feeds Rails and Devise codebases into the model, runs live quizzes showing AI vs real titles, and produces an audio avatar to speak the output — usable and oddly coherent but imperfect, accepted by one CFP and rejected by another.
AI-written summary · about
chapters (18)
- 0:00Introduction and setup
- 1:24Expectations vs reality
- 3:44Talk title reveal and joke
- 4:25Personal background and journey
- 9:02Struggling with ideas and depression
- 11:13First AI-generated quiz
- 13:01Neural networks explained
- 15:12Second AI-generated quiz
- 18:22Third AI-generated quiz
- 22:22Recurrent neural networks for text
- 24:49Training the network on Rails
- 26:16How neural networks learn
- 28:43Training process and loss function
- 30:34Network learning in action
- 31:59Final AI-generated quiz
- 33:03Generating the full talk
- 36:34Talk submission and outcome
- 37:19Q&A and closing thoughts
related talks
30mSteal this talk
21mGPT-3 wrote the description for this talk... Scary or exciting?
21mChalkTalk: Globalizing Data Science Education with AI-generated Videos
59mKeynote: The past, present, and future of AI for application developers
58mKeynote: The past, present, and future of AI for application developers
43mGiving my wife a voice with Ruby and AI cloning (+ discussion)
35mThe Podcast Pipeline Nobody Sees: Behind the Scenes with AI Agents and AI gen
25mFrom Treatise to Tech: The Development of ChatSOC, an AI Research Assistant