ChatGPT here, ChatGPT there… It’s the great digital revolution at the beginning of the year! After OpenIA, many companies have decided to launch their own. If Google recently unveiled Sparrow, Meta is also preparing a revolutionary new tool that has a huge advantage over its peers…
Now remember this name: LLaMA. This innovation is not like ChatGPT or Bing, it is not a system that anyone can chat with. Rather, it’s a tool that Meta says it shares in the hopes of democratizing access to the important and rapidly evolving field of search.
Unlike ChatGPT, LLaMA or Large Language Model Meta AI is therefore a new research tool that will soon help in building Artificial Intelligence (AI) based chatbots. The model developed by META’s FAIR (Fundamental AI Research) team aims to help scientists and engineers explore the applications and functions of AI. It has approved access for specific groups such as universities, NGOs and industrial laboratories.
Meta previously launched similar projects, Galactica and Blender Bot 3, but they quickly fizzled out. In fact, BlenderBot was criticized for simply not being good enough, while Galactica, which was designed to write scientific papers, was scrapped after just three days for producing scientific nonsense.
Why is LLaMA different from other AI-based tools?
As far as META is concerned, the launch of LLAMA represents its commitment to open science – hence the choice to publicly release state-of-the-art fundamental large language models. Also, researchers will have an open resource to further their work. Meta believes they will prove versatile with multiple use cases, as opposed to more elaborate models designed for specific purposes.
Indeed, as the meta explains:
“The LL.M. has shown great promise for more complex tasks such as generating text, conducting conversations, summarizing written material and solving mathematical theorems or predicting protein structures. META is committed to this open research model and we Will make the new model available to the AI research community. »
In his official blog, Meta adds:
“We believe that the entire AI community – academic researchers, civil society, policy makers and industry – needs to work together to develop clear guidelines regarding responsible AI in general and responsible large language models in particular” .
Another distinguishing feature of LLaMA, according to Meta: It requires “much less” computing power than previous offerings and is trained in 20 languages, based on the Latin and Cyrillic alphabets.
Furthermore, with its 13 billion parameters, LLaMA should outperform GPT-3, the model on which ChatGPT is built. Meta attributed LLaMA’s performance to “cleaner” data and “architectural improvements” in the model that improved drive stability.
So it’s not a system you can talk to, but rather a research tool that Meta hopes other AIs will use to solve some of the problems plaguing language models.
This new tool launched by Meta could mark a major development in AI language models!
Source: Meta AI