Microsoft Launches Correction Feature To Overcome AI Writing Errors

JAKARTA Microsoft released the Correction feature for Microsoft Azure customers. Currently, the Artificial Intelligence (AI)-based feature is only available in preview mode as part of the Azure AI Studio service.

This new feature is designed as a security tool that can detect various forms of vulnerability, find 'halusination' problems, and block commands deemed dangerous. This feature was also released to address AI inaccuracies issues.

Microsoft Azure subscribers can use the Correction automatically to detect and rewrite false AI-made content. The Correction system will work immediately once the user activates the feature.

Before correcting the wrong or inaccurate writing according to the user, this system will scan all the writings to find out what parts or text are less accurate. After that, the Correction will compare the writing with the material from the customer's source.

If the error identification process is complete, the Correction will highlight the various errors they found and provide an explanation as to why the writing was wrong. Next, the Correction will carry out its main task, namely rewriting the desired content.

To TechCrunch, a Microsoft spokesperson said that this feature can work accurately because it uses a small language model and a big language model. The combination of this language model can create a output that is in line with the basic documents.

"It is important to note that basic detection does not solve the 'accuracy' problem, but helps align the AI output with basic documents," Microsoft said.