Unconventional programming paradigms for the future, now
summary
Claims that neural nets approximate any function and that gradient descent can outproduce human-written code. Describes layered architectures like VGG16 (convolution + max‑pool + ReLU) trained on GPUs, and image synthesis with pix2pix / GANs that fill sketches into photorealistic cats. Points to a GitHub project called deep neural network evolution using genetic programming to evolve populations (thousands up to ~100,000) of architectures, with some evolved models outperforming human designs.
AI-written summary · about
chapters (4)
related talks
03mRethinking software engineering in the AI era
01h 09mThe Future of Software Engineering
58mPatterns, Predictions and Programming
29mReal Programming
36mFoundational Design Principles and Patterns for Building Robust & Adaptive(…)
46mUniversal Patterns of Evolution
01h 22mWhat's inside the box? Building a deep learning framework from scratch.
01h 04mThe Future of Fast Code: Giving Hardware What It Wants