Especially with ChatGPT developed by OpenAI, a competition to develop a conversational artificial intelligence bot has started between technology companies recently. A new language model that feeds these artificial intelligence bots came from Meta.
Now, Meta, the parent company of Facebook and Instagram, is joining this trend. Meta announced that it has developed a new language model called LLaMA.
LLaMA, unlike others, will be used for more research:
LLaMA (Large Language Model Meta AI) will not be available to everyone, unlike the language model that makes up OpenAI’s ChatGPT. The language model will be entrusted only to the scientific world.
The reason why Meta targets the scientific world is to avoid problems such as the sexist responses we encounter in ChatGPT today. Scientists operating in the field of artificial intelligence will be able to better understand how language models work, paving the way for developing safer language models.
The main reason why Meta does not make this model available to everyone is to prevent the use of malicious responses. The company will only allow academic studies to maintain honesty in use and prevent abuse.
Meta will not be making LLaMA available as a single version either. Different versions of the language model with 7, 33 and 65 billion parameters can be preferred.
So how does LLaMA work?
According to the shared statement, LLaMA takes a set of words as input, just like other large-scale language models, and predicts the next word to generate text. Texts from the 20 most spoken languages were used to develop the language model.
26609
