Many large companies do not allow their employees to use ChatGPT. The reason is that it is wary of OpenAI’s use of the data that users feed into this artificial intelligence (AI). Despite the fact that its creators assure that privacy is guaranteed, the truth is that there is no complete guarantee that our data will be safe from other eyes.
It is no wonder that Microsoft has created an artificial intelligence for the CIA that does not require the user to be connected to the network to use it and which, according to Bloomberg, is based on ChatGPT 4. It should be remembered that the ‘company founded by Bill Gates supplies many material resources to Open AI.
The United States spy agencies thus ensure that their data will be in a safe place, since they use their own and closed data network, independent of the Internet.
Chat GPT’s intelligence is based on OpenAI’s large cloud computing deployment, powered in part by Nvidia’s famous H100 cards. In order to match the computing power normally used by Chat GPT in computing centers connected to the network, Microsoft has spent months adapting a supercomputer located in the state of Iowa.
It is probably the same one that Microsoft has in this state and that was used in 2020 to pre-train Chat GPT 4. Its location was news at the time due to the large water consumption that was needed to cool this powerful machine. It was located in this state because the temperatures are colder than in other areas of the United States.
About 10,000 people are expected to use the data from this spy version of Chat GPT, which is not yet known what it will be called. Its operation is expected to be very similar to the GPTs we can train with standard Chat GPT 4, albeit on a much larger scale. Obviously, this AI is trained, to a large extent, with confidential information.
This class of AI, without the need for an internet connection to be able to interact with it, can make a lot of sense for many institutions and companies, and even for any citizen, as the information is safeguarded from possible prying eyes .
So far, neither the cost of this new infrastructure nor that of its maintenance have been revealed, but both are expected to be high.
Offline data processing is one of the next goals in the deployment of artificial intelligence. Microsoft itself, although it is one of the major participants in OpenAI, is developing an artificial intelligence capable of running on a phone or tablet: Phi 3 Mini, which has a similar capability to that of ChatGPT 3.5.
One of the big advantages of these AIs is that they improve latency, meaning that the responsiveness of an offline AI is much faster than that of an online AI, which can be very important in certain tasks.
Nvidia already has software capable of running artificial intelligence on a PC using some models of its graphics cards. Chat With RTX is the name of Nvidia’s AI capable of running on a PC with a graphics card in the RTX 30 or RTX 40 range, which can be obtained for around 300 euros.
Apple’s latest iPad and Mac releases also appear to be ready to use AI on the same machine, as both the M3 and new M4 processors feature an architecture designed to run Intel applications artificial intelligence These developments will surely be revealed at Apple’s developer conference.
