Microsoft is currently testing the GPT AI language developed by technology brand OpenAI for use in its suite of Office products, including Word, Outlook and PowerPoint.
OpenAI’s intuitive technology products, including Discuss and Dall-E 2, have become Internet sensations due to their ingenuity in creating text and images. Many have speculated on how ChatGPT can be used practically and morally. However, Microsoft is looking to use the company’s AI models more effectively. According to the information.
Microsoft has also tested the functionality of the GPT AI form on PowerPoint and Outlook. These include a feature that allows users to find Outlook search results using AI-powered voice commands instead of keywords in an inbox. Outlook and Word also receive AI models that use suggested email replies or recommended document edits to hone writing skills. It’s currently unknown if this usage will eventually be included in consumer versions of Microsoft Office, or if the brand is simply playing around with the capabilities of the GPT form.
However, this practical use of GPT technology comes after Microsoft invested $1 billion in OpenAI in 2019 and “purchased an exclusive license for the underlying technology behind GPT-3 in 2020,” the post adds.
In addition to its own Office suite, Microsoft may seek to implement the GPT AI model in its Bing search engine in a bid to compete with Google. This may be the most likely product to hit the market, with availability expected in March, according to the brim.
However, OpenAI technology, while excellent, comes with its own set of pitfalls, some of which relate to information accuracy and privacy. The brand’s ChatGPT AI free chat software has been known to fill in information it doesn’t know with incorrect data, which can be a challenge if the form is being developed for a business use case.
In terms of privacy, The Information said Microsoft is developing its own dedicated privacy-preserving models based on GPT-3, as well as GPT-4, which has yet to be released. The company says it has had positive early results in “training large language models on private data,” but has not confirmed whether the model is viable enough for a commercial or even business-class product.