Pause AI development is “stop the wind with your hands”

“Twe are at this moment a phenomenon that has left us all surprised, perplexed, I would even say”, that there was an artificial intelligence model, Fable, from Anthropic, “which, overnight, without any warning, the button was suddenly turned off”, he points.

As a matter of fact, “I was doing a project with Fable to understand the capabilities of the model, and suddenly I read a report at Bloomberg that it had disappeared and, in fact, “it was not already there,” reports Paulo Dimas, considering “awesome” that a “superpower” such as the US has the power to “disconnect artificial intelligence.”.

In mid-June, Anthropic suspended public access to the most advanced AI models to comply with an export control directive that obliges it to prevent access to the service by foreigners for national security reasons, but now, in early July, Washington lifted the restrictions.

According to Paulo Dimas, there is now “a very broad set of artificial intelligence models, many of them developed in China, those in the leadership, continue to be developed.”.

“ We will achieve the capabilities of this model that at the moment” is only available to a set of companies or is available on request, we will have in open models developed by China”, so take a break “is to stop the wind with your hands”, he says.

In fact, “how are we going to prevent these technological advances by these Chinese companies because, at the end of the day, what interests us is then the application of these models. And, therefore, I think [...] it makes no sense”, insists Paulo Dimas.

Now, “what makes sense is that we have models that are responsible from the point of view of the use of artificial intelligence”, that is, “models that we can understand what the alignment they have, [...] morality, to use the term of the Pope, which they have, that is what is important, because we believe these will be the most competitive”, stresses the CEO to the Responsible AI Center (name in Portuguese).

On the other hand, at European level, “we have to collaborate strategically in the creation of these large-scale models”, he argues.

The way to do this is, “again, as it has been done,” he says, using the examples of the models of AI EuroLLM, Amália, “which is through collaboration, through openness”.

China, in a way, “is doing this through very funded startups, we in Europe do not yet have that agility of startup funding and we do not have the scale, but I think we can have this alignment,” says Paulo Dimas.

“ I do not think that this is impossible, [...] the European Commission and the European Innovation Council also have an important role to play in aligning Europe with this absolutely strategic plan of sovereignty in artificial intelligence,” he continues.

“ We are perhaps building two, three towers of Babel centered in which power is in the hands of few and we must evolve to a collaborative model in which we can all participate in the creation of artificial intelligence ", insists Paulo Dimas, referring to open models.

Because when “we work on open models, we are sharing what we are doing, at the bottom, each of us is making a contribution to the greater good”, he adds.