DeepMind AlphaCode Achieves Median Rank in Competitive Programming
DeepMind's AlphaCode system writes computer programs at a competitive level, placing within the top 54% of participants on Codeforces. The AI system uses massive-scale code generation and smart filtering to solve novel problems.
DeepMind introduces AlphaCode, an artificial intelligence system that writes computer programs at a competitive level. The system achieves an estimated rank within the top 54% of participants in programming competitions by successfully solving new, unforeseen problems that require critical thinking, logic, algorithms, and natural language understanding.
The technology relies on transformer-based language models to generate code at an unprecedented scale and then intelligently filters the output down to a small set of promising programs. DeepMind validates this performance using 10 recent competitions on Codeforces, a popular platform that attracts tens of thousands of human programmers from around the world.
By performing at about the level of the median competitor on these challenges, AlphaCode marks a major milestone as the first AI code generation system to reach a competitive level in programming contests. To encourage further innovation in problem solving, the research team releases a new dataset of competitive programming problems and solutions on GitHub.