Microsoft provides a large model of GPT to the US government. How to ensure security?

Microsoft published a blog on June 7, announcing that it would provide OpenAI artificial intelligence large models to U.S. government agencies, leveraging the capabilities of its Azure cloud services. This is the first time that Microsoft has introduced GPT technology into a government agency, and it is also the first case in the world where GPT represents a large AI model and is introduced into the government. This move sends an unusual signal, especially in the context that institutions and large enterprises around the world have banned large GPT models on the grounds of "security".

Azure government customers can use the Microsoft Azure OpenAI service to take advantage of OpenAI's two large language models, the latest GPT-4 and the earlier GPT-3 large model. Microsoft said federal, state and local government customers can use the two models to complete a variety of services, such as generating answers to research questions, generating computer code and summarizing field reports. Additionally, language models can be tuned for specific tasks, including content generation, language-to-code translation, and summarization.

Microsoft hosts OpenAI models in its commercial cloud computing space, separate from the cloud used by Azure government customers. The latter adheres to various specific security and data compliance rules, and data from Azure Government customers will not be used to train the model.

OpenAI is the development company of the chat robot ChatGPT. It has products such as the large language model GPT-3 and the multi-modal model GPT-4. Microsoft is OpenAI's largest investor, investing US$1 billion in it in 2019, and added a second multi-year investment in January 2023, with the investment reportedly reaching US$10 billion.

Since OpenAI launched ChatGPT, large AI models have ushered in an unrivaled wave of development. Large companies including Google, Alibaba, and Baidu have released their own large AI models. Among them, Microsoft and Google have successively launched AI products in the field of network security, which have attracted widespread attention and discussion in the industry.

Microsoft has made OpenAI models available to its commercial customers, and the Azure OpenAI service is growing rapidly. As of May this year, it has 4,500 customers, including groups such as Volvo, IKEA, Mercedes-Benz Group and Shell.

In today's world, artificial intelligence technology is developing rapidly, and large-scale language models, as an important part of it, have broad application prospects. In this context, Microsoft's move is regarded as a new exploration of the development of AI technology.

However, as large-scale language models gradually mature and are applied, the security issues they bring cannot be ignored. Relevant experts pointed out that the possible risks of large language models include privacy leakage, human value deviation, misleading information generation and other aspects. In response to these issues, further research and exploration are currently needed to ensure the safety and reliability of artificial intelligence technology.

In addition to security issues, the high cost of large language models is also a factor restricting the popularity of their applications. Currently, the training and running costs of large language models are very high, which also limits their application in some fields. Therefore, in the future development process, how to improve the efficiency of the model and reduce the cost will become an important research direction.

In general, Microsoft's move to use Azure cloud services to provide large OpenAI artificial intelligence models to government agencies has brought new explorations and opportunities to the development of artificial intelligence technology. At the same time, in the process of applying this technology, we also need to pay close attention to issues such as safety and cost to ensure that it can bring more value and benefits to society.

The above is the detailed content of Microsoft provides a large model of GPT to the US government. How to ensure security?. For more information, please follow other related articles on the PHP Chinese website!