DeepMind’s AI lab unveiled AlphaCode, a tool capable of writing code autonomously. According to the developers, the algorithm is able to compete with mid-level programmers.
Introducing #AlphaCode: a system that can compete at average human level in competitive coding competitions like @codeforces. An exciting leap in AI problem-solving capabilities, combining many advances in machine learning!
Read more: https://t.co/yaXfMWtEfe 1/ pic.twitter.com/Mvhc6Jm7E0
— DeepMind (@DeepMind) February 2, 2022
The AI’s effectiveness was tested on the Codeforces platform, which daily offers tasks for programmers requiring critical thinking, logic, and knowledge of algorithms.
The developers loaded 10 tasks in English into AlphaCode. The same tasks were solved by 5,000 Codeforces users.
The tool generated several solutions for each task and tested the code’s functionality on its own. As a result, AlphaCode’s answers would have placed it among the top 54% of participants in the contest, and among the top 28% of Codeforces users over the past six months.
In DeepMind, the current skill set of AlphaCode is applicable only in competitions. However, its capabilities open the door to the creation of future tools that may one day automate programming completely.
According to Codeforces founder Mike Mirzayanov, AlphaCode exceeded all his expectations.
“I was skeptical, because even in simple competitive tasks it often requires not only to implement the algorithm, but also — and this is the hardest part — to invent it,” he said.
Back in May 2021, Microsoft introduced a tool for autocompleting code based on the GPT-3 language model.
In June, GitHub released the AI assistant Copilot, helping programmers write code. Four months after the release, the company said that 30% of new code on the platform was created with the tool.
In August, the OpenAI AI lab introduced the Codex, which translates natural language into code.
Subscribe to ForkLog news on Telegram: ForkLog AI — all the news from the world of AI!