Meta wants to really compete with OpenAI and Google: it will launch its own paid GPT-4, according to FT

In recent months we have gotten used to living with tools like ChatGPT, Bard and Bing Chat. Whether we use them or not, they are present in some way or another in our lives. Now, we may have to add another one to the list soon.

Meta, the parent company of Facebook, Instagram and WhatsApp, does not want to be left out of the artificial intelligence (AI) race. And while it has invested millions of dollars in the metaverse, it’s also making a big bet on the world of generative algorithms.

The imminent launch of a new Meta model

Meta’s experience in the field of AI is not recent. For years the company led by Mark Zuckerberg has used technologies derived from this discipline to make it work part of your social networks, such as the advertising system and recommendation algorithms.

In addition, for years their research laboratories have been working with models. This is why it didn’t take long for him to respond when OpenAI fueled the current AI boom with the advent of GPT 3.5, the engine that powered the viral chatbot ChatGPT:

Three months later, Meta launched an equivalent proposal called Calls. A big language model trained with millions of parameters. The truth is that, despite the fact that it ended up leaking, LLaMA was intended for the field of research, not for the general public.

This is what is about to change. at least according to Financial Times, which requires according to one of his sources that the launch of the evolution of LLaMA is “imminent”. This time, however, it will not be limited to researchers or academics, but will have a broader deployment.

When this presumable new model arrives, it will do so with a commercial focus. This means that those who use it will have greater freedom to use its outputs (data output, for example, a written text). For example, for commercial purposes and not just research.

The latest from Meta in terms of AI, whose name is still unknown, will arrive with three features that will allow it to differentiate itself from GPT-4, the most recent model of OpenAI. One of them is that it will be open source, with all the advantages and disadvantages that this presents.

Being open source, any person or entity with the necessary resources will be able to know the details of its operation. This does not happen in the latest models of the company led by Sam Altman, whose code is fully controlled by its creator.

Without referring to a particular model, Meta has said in the past that the idea of ​​releasing an open source model may be an avenue to explore concerns surrounding AI. Let’s remember the letter that Elon Musk, among others, signed to stop its development.

The fact that an advanced model is open source may mean that some people use its capabilities with illegal or immoral purposes. Among them we find the creation of pieces that promote misinformation, among other activities.

Another differential aspect of the successor to LLaMA will be its versatility for companies. The company will provide an option, possibly for a fee, so that companies can customize the model with their own data sets to better meet their needs.

We do not know at the moment when it could arrive. If we stick to the term “imminent”, it could happen whenever. It’s time to wait to learn more about Meta’s next move and how the race for artificial intelligence becomes more and more intense.

Images: Goal

In Xataka: OpenAI has already announced its strategy to prevent risks with superintelligence: superalignment