What Is in Zuckerberg’s Essay on AI

Mark Zuckerberg, Chief Executive Officer (CEO) of the Meta Platform, parent company of Facebook, has recently written a 6,500-word essay on Artificial Intelligence (AI), drawing much attention from industry experts. Tech experts and many high-profile tech executives write articles often; however, Zuckerberg’s essay is very unusual and uncommon. In the essay, he has covered some key areas, which, among others, include the AI race, its negligence to society and the US government’s role in regulating and promoting technology. Additionally, the comparison between open-source AI models and open-weight AI models has come under elaboration. 
As soon as AI started rolling out through widespread application and massive investment, creating a boom in this sector, two concepts, open AI models and open-weight AI models, have emerged in the AI industry. Zuckerberg is the very high-profile tech executive who has first tried to draw equivalence between these two AI concepts. He is believed to have resorted to such a long write-up on AI with an intent to justify open-source AI, placing this application in the same status as open-weight models, because his company, Meta, is going to release open-source models soon. In the essay, Zuckerberg has mentioned that Meta was strongly supportive of open source, including open-source AI models.

Zuckerberg’s essay has focused on open-source AI models and open-weight AI models. Although both the concepts are being widely discussed in the AI industry, there is a difference between the two. Open-source AI models refer to the source codes which are available in complete form, and anyone can easily take, use or modify them. The common people using the application can recommend necessary changes or updates in order to make it better. If the company which offers open-source AI modes accepts those changes/updates, they can incorporate them into the code. On the other hand, open-weight AI models, in question, are not open at all and do not serve the same purpose as open-source AI models do. The company, which offers open-weight AI models, prepares a model using methods that are not disclosed to its users. As opposed to open-source models, users can neither find the code used to train the model nor retrain it themselves or add data. The level of standard and intelligence built into the open-weight AI model is inevitably set by the people who have trained the model. Weights are, in fact, the result of training – numerical values that influence how the AI’s digital brain works.
Since modification can be done and found useful in open-source AI models, the users can tailor weights for their purposes with a view to making an AI model specialise in one particular area. As discussed in the essay, modification of weights can make a model more useful and prone to adjustment. However, there is a bad side too, as modification sometimes may be harmful. If bad people with substantial computing resources get access to change or remove weights, which blocks the AI from carrying out cyberattacks or giving instructions on how to make detrimental objects. In a nutshell, open-source AI models are categorised as the products of the community developers and open for the whole world, while “open-weight AI models are closely tied to the company which creates and promotes those models.
In the essay, while elaborating on open-source AI models and open-weight AI models, Zukerberg has also highlighted Meta’s position in adopting the models. Zuckerberg’s essay has clearly explained that Meta decides when it trains the next generation of its open-weight AI models, and its capabilities – not those of the community – determine the models’ basic level of intelligence. In consequence, Meta is believed to remain in an advantageous position compared to other competitors to reap benefit from AI models. This is because of the company’s familiarity with how they are developed and their willingness to invest in the computing power it takes to deploy them to users. The essay states that the balance of power in open-weight AI models is much more centralised.
In the essay, Zuckerberg has written that one of the US’s current disadvantages in its race against China to develop superintelligence systems is the difficulty of building infrastructure stateside. It is learnt that Meta has contemplation of investing $145 billion this year and $600 billion by 2028 for infrastructure development, particularly building data centres. Social resistance against building data centres is also considered a threat to this sector, and New York City’s ban on building data centres for one year is a good example of this context. So, Meta has undertaken some community development programmes, and they have already allocated one billion dollars for this purpose, which is also the focal point of this essay. 
Zuckerberg, in his essay, has conspicuously mentioned a world where “open-source AI models” allow anyone and everyone to harness AI’s power and bend it to their own purposes. He also added that there may be a grain of truth in that, but it is also self-serving. Zuckerberg has written a 6,500-word essay on AI, when discussion on open-source AI models and open-weight AI models has become a big debate. Some experts and industry leaders are viewing open-source AI models as a threat to the whole AI industry. Even the US Government is going to regulate AI’s role. Whether Zuckerberg’s long, easy will change the course of AI growth or ensure coexistence of two AI concepts is not made clear to the industry experts. Perhaps the next few months will give some idea about the impact of this long essay. However, one key point from Zukerberg’s long essay is very clear: that open-source AI models is going to coexist with open-weight AI models in the AI industry.
___________________________________________
The writer is a certified anti-money laundering specialist and banker based in Toronto, Canada. He can be reached at [email protected]