
According to a report from Bloomberg, Microsoft has spent hundreds of millions of dollars building a giant supercomputer to power OpenAI's ChatGPT chatbot.
Microsoft owns a super server system that operates ChatGPT
Microsoft revealed that the company has connected tens of thousands of Nvidia A100 chips and rack servers to build hardware for ChatGPT and Bing AI bots.
In blog posts, Microsoft explains how to create Azure's artificial intelligence infrastructure used by OpenAI, and the company also reveals why processing systems are getting more powerful.
To build the supercomputer powering OpenAI's projects, Microsoft linked thousands of Nvidia graphics cards (GPUs) to the Azure cloud computing platform, allowing OpenAI to train increasingly powerful models powerful and unlocking the limitless potential of AI on ChatGPT and Bing.
According to Bloomberg, Scott Guthrie, Microsoft's vice president of AI and cloud, said the company has spent hundreds of millions of dollars on the project.
Although this is only a modest number for Microsoft, it also shows that the technology corporation has recently expanded its multi-billion dollar investment over many years in OpenAI, ready to pour more money into the AI market.
Microsoft has been working hard to optimize Azure's AI capabilities with the launch of new virtual machines using Nvidia's H100 and A100 Tensor Core GPUs, as well as the Quantum-2 InfiniBand network, which was unveiled in November. 2022
According to Microsoft, this direction will allow OpenAI and other companies that rely on Azure to train larger and more complex AI models.
OpenAI announces GPT-4 to help elevate ChatGPT
OpenAI has officially released GPT-4, an upgraded language model to help ChatGPT work more accurately.
OpenAI calls GPT-4 "the latest milestone in the effort to scale deep learning" and claims it has demonstrated human-level performance across a variety of professional and academic standards.
OpenAI says it has rebuilt its deep learning platform over the past two years and worked with Microsoft Azure to design a supercomputer for this massive workload. GPT-3.5 is the first test run of this system.
Although the new system is more accurate than its predecessor, it still has actual errors. With one of the examples OpenAI provides, it calls Elvis Presley an actor's son. GPT-4 also lacks knowledge of events after September 2021 and cannot learn from it.
However, OpenAI has improved many of the security features of GPT-4, including an 82% reduction in the model's tendency to respond to unauthorized content elements compared to GPT-3.5, and it only responds to about 29% of sensitive requests such as medical advice and self-harm.
OpenAI wants GPT-4 to be a valuable tool in improving people's lives by powering many applications.
The company also acknowledges that they still have a lot of work to do and wants to improve this model through a joint effort of the community on the basis of "building, discovering and contributing" to the model together.
At the current stage, only ChatGPT Plus members paying a monthly fee of 20 USD can get the experience of entering text from images of GPT-4 images through ChatGPT, while regular users have to wait a little longer. People can also experience ChatGPT and GPT-4 in Microsoft Bing and the Edge browser the company announced in February.