Xinhua Finance, Beijing, December 25th (Bu Xiaoming) Reuters reported on the 23rd that Microsoft's AI assistant 365 Copilot will add third-party AI models to reduce dependence on OpenAI technology.
Microsoft and OpenAI have a close relationship. The former is the major "benefactor" of the latter, while the latter prioritizes providing large language model support for the former. Reuters reported that Microsoft is seeking to reduce 365 Copilot's reliance on OpenAI, primarily to lower costs and meet the speed demands of enterprise users.
When Microsoft released 365 Copilot in March 2023, its key marketing selling point was the integration of OpenAI's ChatGPT-4 model. According to the promotion, 365 Copilot is an AI assisted productivity tool that combines big language models, business data, and Microsoft 365 programs to automatically generate documents, emails, and PowerPoint presentations, aiming to enhance users' work efficiency and creativity.
However, the actual performance of 365 Copilot did not impress customers and Microsoft insiders. The US website Business Insider previously reported from anonymous Microsoft employees that 365 Copilot is "terrible" and "fancy", with "75% of the time running poorly" during use. The report also cited a survey conducted by Gartner Consulting on 123 IT industry executives, which showed that only 4 respondents believed that 365 Copilot provided "significant value" to their companies.
Reuters reported that Microsoft is training smaller models including Phi-4 and customizing other open weight models to make 365 Copilot faster and more efficient. The report states that Microsoft CEO Satya Nadella is closely monitoring the progress.
Outside of 365 Copilot, Microsoft's other businesses are also reducing their reliance on OpenAI big models. For example, GitHub, which Microsoft acquired in 2018, added large models developed by Anthropic and Google in October this year to replace ChatGPT. The Microsoft chatbot "Copilot" now uses both an internal big model and an OpenAI big model.
The Wall Street Journal reported on the 20th that due to a lack of training data and extremely expensive research and development, it is unclear when or if the upgraded product ChatGPT-5 will be launched. ChatGPT-5 has been trained for over 18 months.
The report states that investors valued OpenAI at $157 billion in October, mainly due to OpenAI co-founder Sam Altman's promise that ChatGPT-5 will demonstrate "significant and leapfrog progress" in all tasks. But now the entire industry is discussing whether the overall development of AI in the United States has entered a bottleneck period.
OpenAI co-founder Ilya Sutskwell recently said, "We've run out of data, there's no more data left... Data is the fossil fuel for AI