Microsoft considering ChatGPT in Office apps

Rumours say Microsoft is already using it in Word

Microsoft is said to be working to include OpenAI's ChatGPT chatbot technology into some Office applications.

According to The Information, Outlook, Word and PowerPoint users will be able to use the technology to produce content automatically using natural language prompts.

'For more than a year, Microsoft's engineers and researchers have worked to create personalised AI tools for composing emails and documents by applying OpenAI's machine-learning models to customers' private data', said a person with direct knowledge of the plan.

'Engineers are developing methods to train these models on the customer data without it leaking to other customers or falling into the hands of bad actors', they added.

ChatGPT is a chatbot that can answer a wide variety of questions while emulating human speaking styles. OpenAI launched it for free public testing on 30th November and it has generated excitement with its human-like responses to natural language questions.

People have already used the app to write essays, poems, lyrics, cover letters and CVs. It is also capable of solving mathematical problems and analysing computer code. However, its outputs are not fully trustworthy and some insitutions - including the International Conference on Machine Learning, Stack Overflow and public schools in New York City - have banned submissions using its outputs.

OpenAI has admitted that ChatGPT may sometimes provide erroneous information.

The Information claims Microsoft is already using OpenAI's text-generating GPT model to enhance autocomplete functionality in Word. It is also using the technology to improve Outlook's search results.

Just last week, a report claimed that Microsoft is working to incorporate OpenAI's ChatGPT chatbot to its Bing search engine to make it smarter and to challenge Google's search dominance.

Microsoft could roll out the new feature before the end of March, the report added.

Microsoft AI tool can simulate human voice with three-second sample

In related Microsoft AI news, company researchers have announced a new text-to-speech (TTS) AI model named VALL-E, which can precisely replicate a person's voice using just three-seconds of audio.

Microsoft describes VALL-E as an AI-driven TTS system that can be taught to talk like anybody. Researchers call the tool a "neural codec language model".

Once VALL-E has learned a particular voice, it can simulate that person saying anything else, and does so in a manner that preserves the speaker's emotional tone.

VALL-E is based on EnCodec, a technology Meta introduced in October 2022. It produces distinct audio codec codes from text and acoustic prompts, in contrast to conventional TTS systems that usually generate speech by modifying waveforms.

The tool analyses a person's speech and breaks it down into distinct parts (tokens). It then uses training data to determine how that voice would sound if it said sentences other than the three-second sample.

The AI tool has been trained with 60,000 hours' worth of audio input from thousands of different people, including audio books in the public domain.

Its developers say VALL-E could be combined with other generative AI models like GPT-3 to create audio content. It could be used for high-quality text-to-speech applications, as well as speech editing applications that allow a text transcript to be used for editing and changing a recording of a person.

However, as with any human-simulating technology, there are obvious security implications that must be considered early on.