Is the programming cord created by artificial intelligence (AI) perfect?
This concern is increasing due to AI that recently learns data incorrectly or false results with errors.
Microsoft (MS) shows a development tool for automatically reviewing the code that AI writes to reduce the user’s concern.
According to the US Zedinet, Microsoft (MS) released the artificial intelligence (AI) based code review automation tool, “JIGSAW”.
Jigsaw is a supporting tool that automates the programming code creation process. The programming code that you created is compiled, including whether to compile whether the error message is resolved and the code is integrated to test whether you create the desired result.
The reason for developing jigsi from Microsoft is because the code created by AI is not perfect.
Recently, a service that utilizes AI language models such as GPT-3 to understand the natural language, and the service to switch to source code is being released. Codex and Power Fx are representative.
With these services, you can develop applications by communicating with AI as natural language as you talk. Non-developers can also participate in development, and professional developers can reduce the troublesome coding work and increase productivity.
However, the user is ambiguous and the AI may not work properly because the AI is misunderstood, or to switching the words, and is not working properly.
An error may occur during the coding process. AI is because there is a mistake that is based on the probability, and because it may include erroneous data in the learning process.
Microsoft has introduced an input / output value comparison function to ensure the quality of the code created by AI. It is a function to determine whether the input value and the output value are connected and output as intended.
If the user enters the description of the description of the contents of the code in the code, and presents exemplary I / O data, the jigsaw reviews the code automatically based on the contents.
In addition, after review, the output is transmitted to the person in charge for the final confirmation of confirmation and editing. When a feedback is caused by an error in the output result, the code automatically improves the code through the recovery process.
The Microsoft Research team improved the output accuracy to more than 60% through jigsia, and if you introduce user feedback, you can bring up the accuracy to more than 80%.
Currently, Jigsuk can be used in GPT-3, Codex, Copilot, and language only supports Python.
Microsoft’s Raúl Sharman Senior Researcher said, “I believe that it can help to improve the productivity of developers through jigsi,” he said, “He said,” he said.