DeepMind AlphaCode AI Rivals Human Programmers, Surpasses GPT-3
DeepMind introduces AlphaCode, an AI programmer that ranks in the top 54% of human coding competitions and outperforms OpenAI's Codex. The system translates natural language into functional code across multiple programming languages.
Google-backed DeepMind unveils AlphaCode, a new artificial intelligence system that writes computer code and outperforms OpenAI’s GPT-3-powered Codex tool. The AI ranks in the top 54% against over 5,000 human participants across 10 Codeforces programming contests, marking the first time an AI reaches this competitive level. AlphaCode translates plain language instructions into accurate, ready-to-use computer code.
The system relies on a massive 41.4 billion parameters trained on public GitHub databases, enabling it to generate functional code in nearly a dozen programming languages. DeepMind researchers highlight that while Codex excels at completing simple Python functions, AlphaCode pushes boundaries by solving complex, full-scope competitive programming problems that involve much more intricate task specifications.
Despite these impressive capabilities, AlphaCode remains very new and faces a long path before widespread public availability. Meanwhile, OpenAI currently holds a practical advantage with its publicly accessible Codex API. If AI coding tools like AlphaCode eventually become mainstream, they promise to significantly reduce development time and costs, freeing human developers to focus on the more creative aspects of software design.