AI can quickly write code based on common languages

  • by
  • 3 min read

Last few years, Researchers have used artificial intelligence it Improve translation Between programming languages ​​or mechanically Solve the problem. For instance, the unreal intelligence system DrRepair has been confirmed to unravel most issues that produce error messages. But some researchers dream of someday that AI can write applications based on easy descriptions from non-experts.

Tuesday, Microsoft with OpenAI Shared plans to introduce GPT-3 (probably the most superior textual content technology fashions on the earth) into programming based on pure language descriptions. This is the primary industrial utility of GPT-3 since Microsoft invested $1 billion in OpenAI final yr and obtained the unique license for GPT-3.

Microsoft CEO Satya Nadella (Satya Nadella) mentioned in a keynote speech on the firm’s Build Developer Conference: “If you utilize pure language to explain what you need to do, GPT-3 will offer you essentially the most related An inventory of formulation so that you can select from.” “Write your personal code.”

Provided by Microsoft

Microsoft vice president Charles Lamanna (Charles Lamanna) told Wired that the advanced technology provided by GPT-3 can help people deal with complex challenges and empower people with little coding experience. GPT-3 will convert natural language into PowerFx, which is a fairly simple programming language similar to the Excel command Microsoft introduced in March.

This is the newest demonstration of making use of AI to coding. Last yr at Microsoft Build, OpenAI CEO Sam Altman Downgrade A language mannequin that has been fine-tuned by code in GitHub and mechanically generates Python code strains.As WIRED detailed final month, startups like SourceAI are additionally utilizing GPT-3 generated code. IBM demonstrated last month how its Project CodeNet (with 14 million code examples from more than 50 programming languages) can reduce the time required to update millions of lines of Java code for a car company from one year to one month.

Microsoft’s new features are based on Neural Networks The architecture called Transformer used by large companies includes Baidu,, Google,Microsoft, Nvidia, And Salesforce uses text training data crawled from the Internet to create large-scale language models. These language models continue to expand. The largest version of Google BERT is a language model, which was released in 2018 and has 340 million parameters and is part of a neural network. The GPT-3 released a year ago has 175 billion parameters.

However, such efforts nonetheless have an extended option to go. In a current take a look at, the perfect mannequin was efficiently accomplished solely 14% of the time in an introductory programming problem written by a gaggle of AI researchers.