Keynote: Writing Code? Pfft... Evolve it Instead!
summary
Shows a working pipeline that uses genetic algorithms to evolve code strings into runnable programs by targeting the program output with a fitness function, using population, crossover and mutation and roulette-wheel selection to avoid premature convergence. It uses a Brainfuck-like eight-character language (Turing-complete) so random genomes can become valid programs, adds timeouts and memory bounds to discard nonterminating or out-of-memory runs, and finds simple programs (greeting, add inputs, Fibonacci) but convergence is slow so full developer replacement is not imminent.
AI-written summary · about
chapters (9)
- 0:00Introduction and background
- 1:53Technology displacing jobs
- 8:15Genetic algorithms explained
- 13:46Evolving programs with Brainfuck
- 25:36Will software jobs be automated?
- 31:36Historical job displacement
- 35:58Future of automation debate
- 38:48Societal responses to automation
- 40:55Q&A session
related talks
33mLinear Genetic Programming in Python Bytecode
40mGenetic Programming in Clojure
46mTight Genes: Intro to Genetic Algorithms
09mOuvroir de Code Potentiel
48mTight Genes: Intro to Genetic Algorithms
15mKeynote: Pre-training and Fine-tuning of Code Generation Models
01h 32mCan AI replace programmers?
26mThe present and future of programming with AI (Keynote)