In the world of artificial intelligence, a new language model developed by Meta, Facebook’s parent company, is giving ChatGPT, OpenAI’s popular chatbot, a run for its money. Meta recently unveiled LLaMA, a program capable of generating well-constructed sentences, answering questions, and even translating and summarizing texts. One of the things that sets LLaMA apart from ChatGPT is its smaller size and memory usage: it is between 2.6 and 26 times smaller than ChatGPT’s GPT-3 model but performs just as well, if not better.
However, what really caused a stir in the AI world was Meta’s decision to make LLaMA open source. The company not only published the details of the algorithm’s recipe but also shared the precise quantities of ingredients to be used in the training phase of language models. This move is a significant departure from the norm, as companies such as OpenAI, Apple, Amazon, and Google typically guard this information very closely.
Since LLaMA’s release, other companies have developed their own versions of the model. Hugging Face, for example, has created over 30 derivatives of different versions of Meta’s model, and companies such as EleutherAI and the Technology Innovation Institute of the United Arab Emirates have also developed their own models. These models are smaller and less computationally intensive to use than ChatGPT’s GPT-3 and GPT-4, making them more accessible to individual users.
In fact, computer scientist Artem Andreenko demonstrated that LLaMA can be run on a Raspberry Pi single-board computer, a small and affordable device popular among DIY enthusiasts. Additionally, Nomic has developed Gpt4All, a software program that allows users to install several models derived from LLaMA on their desktop computers and use them offline.
This development has significant implications for the future of conversational agents and AI in general. With LLaMA and other smaller models available to individual users, there may be less reluctance to use conversational agents, which are currently mostly online. As more companies adopt the open-source model, the AI landscape could be redefined, with greater access to powerful language models and greater potential for innovation.
