On yesterday’s Italian evening, Thursday 18 January, Mark Zuckerberg had shared a video post on Facebook to provide “some updates on our efforts in the field of artificial intelligence.”
“Ours” meaning from Meta, his company: “The long-term plan is to create Strong AI, make it open source in a responsible way and make it widely available so that everyone can benefit from it”. What do these words mean? Let’s see them in detail.
Science and science fiction 40 years after Terminator, James Cameron and the fear of AI: “I warned you, you didn’t listen to me” by Emanuele Capone 31 July 2023
What is Strong AI?
Meanwhile, what is it what Zuckerberg called “general intelligence” and which can be translated into Italian as IA Forte? It is what is known in the industry as breaking latest news, i.e. Artificial General Intelligence and, as we have often explained on Italian Techis an AI capable of solving general problems on its own, without limiting itself to the specific ones for which it was programmed.
It is not the sentient AI imagined by films like Terminator or (more recently) the pleasant The Creator, although it is often called that, but somehow yes. And Meta is not the only one aiming to create one: OpenAI also announced the same programs in March 2023, sparking a lot of perplexities and fears in the scientific community.
Why it is important that it is open source
At the time, the doubts were mainly linked to the fact that (despite the name) OpenAI is very closed, closely linked to Microsoft for understandable economic interests: little or nothing is known about how ChatGPT was developed and how it is trained, about the datasets on which it learns, about the sources of its knowledge. And the same will presumably be true for the possible Strong AI which will come out of Sam Altman’s company.
Zuckerberg has though chose to follow a different path: “We will make it open source,” you can hear him say in the video accompanying the post, which is an important difference. LLaMA 3, the language model from which Meta’s (possible) Strong AI will be born, is one of the few that can be considered open source among those used by large companies.
Because this point is important? Because knowing how AIs are trained, what their biases may be, if, how and by whom their training was done alignment (things?), possibly also being able to intervene from the outside, is crucial for correcting any errors. That could have catastrophic consequencesin the case of a Strong AI.
Artificial intelligence Stealing data from ChatGPT, using ChatGPT: this is how AI reveals names, surnames, faces and addresses of people by Emanuele Capone 03 December 2023
Investments of almost 15 billion
In the post, Zuckerberg said and wrote something else: “We are building a huge computing infrastructure to support this future roadmap, including 350 thousand H100s by the end of this year”, for an overall computational capacity of “almost 600 thousand H100 equivalent if you include our other GPUs.”
What does it mean? The acronym GPU stands for graphic processing unit, i.e. graphics processing unit or graphics processor: to simplify, it is the video card usually used in gaming computers, capable of performing complex or very complex mathematical calculations at very high speed. Not just for video editing or rendering typical of video games but also (and this is the point that interests Zuckerberg) for the activities of machine learningin which similar mathematical operations are performed on huge datasets.
At the moment, the market leader, which practically everyone who wants to develop AI turns to, is Nvidia (here a portrait of the CEO): the H100s are its flagship graphics cards in the professional world, and each costs around 40 thousand dollars, a little more or a little less. And Meta is thinking of investing around 14.5 billion dollars to buy 350 thousand before the end of the year.
@capoema