Another significant barrier is privacy. Microsoft will have to tailor its models to specific users without jeopardising their data. According to The Information, Microsoft has been working on privacy-preserving models based on GPT-3 and the yet-to-be-released GPT-4. According to reports, Microsoft researchers have had early success in training huge language models using private data.
These models might be used by Microsoft to collect and summarise information from Teams Meeting transcripts, and then add pictures created by OpenAI’s Dall-E 2 image-generating model to PowerPoint presentations. Researchers are said to have shown their Office integration work to Microsoft CEO Satya Nadella, although it’s unclear whether or not these GPT- or Dall-E 2-powered models will be accessible in Office products.
