Microsoft says its new tool can help with AI hallucinations

The new tool works by first flagging the text that has some sort of factual error using “Groundedness Detection”, a feature that according to Microsoft “identifies ungrounded or hallucinated content”. It then fact-checks it by comparing it with a correct source, which can either be a document or uploaded transcripts.

Introduced earlier this year in March, Groundedness Detection is similar to Google’s implementation in Vertex AI, which also lets customers ground models using third-party providers, datasets or Google Search.

Available as part of Microsoft’s Azure AI Content Safety API, which is currently in preview, Correction can be used with other AI models like Meta Llama and OpenAI GPT-4o. However, experts have said these tools like these cannot help address the root cause of hallucinations.

According to a recent report by TechCrunch, Os Keyes, a PhD candidate studying at the University of Washington says that while this approach can help reduce some problems, it will generate new ones.

Last month, Microsoft introduced a new Copilot feature that summarises Word documents, making it easier to go through lengthy documents.

From the homepage

In first speech after Karur tragedy, Vijay lays out his ideology, equates DMK with ‘syndicate of dacoits’

Rethink on Chandigarh Bill after Punjab protests, Govt says final decision not taken

With battle lines clear for 2026, Nitish’s 10th return as CM a reckoning for INDIA amid drift