Meta will soon take a more mainstream approach to AI development, with the upcoming commercialization of LLaMA.
If we have heard a lot about ChatGPT and Google Bard in recent months, Meta is also particularly invested in AI. We thus knew for many months and the leak of LLaMA that the firm of Mark Zuckerberg already had its own language model. First shared with a limited audience, it could quickly become mainstream.
LLaMA for all
What do we know about LLaMA, Meta’s language model? For the majority of the public, the answer would be: not much. And this, despite some somewhat unusual news, such as the one explaining that he would flirt less well than GPT-4. This relative secrecy is wanted by the Californian giant. He presented the language model last February and limited its access to researchers and academics.
But things are about to change, according to several insider sources who spoke with the FinancialTimes. Indeed, Meta wants to move to a new stage and market LLaMA to companies. This opening to the economic world should, according to the newspaper, be done very soon.
Customizable AI
This shift should be all the more interesting as LLaMA will be provided to Meta customers in an open source version. They will thus be able, thanks to access to the data and the codes used, to adapt artificial intelligence to their own needs. In particular, they will be able to build software from the LLaMA language model.
Generally speaking, Meta takes a deliberately different approach than competitors like DeepMind (Google) or OpenAI (Microsoft). For Yann Le Cun, who heads AI operations at Meta, open source platforms developing artificial intelligence should very quickly be as efficient as those remaining in the secret in the coming months. It would therefore be smarter for Meta to take the lead, at the risk of its language model becoming obsolete before it was even popular.
Source : The Verge
2