Mark Zuckerberg, sporting a new bomber jacket and shedding his metaverse shackles, has unveiled Meta's latest offering: Llama 3, a free and open-source large language model (LLM). This move comes as no surprise, as Meta, like many other tech giants, has been playing catch-up with OpenAI's incredible expansion. While the metaverse may still come to fruition in some form down the line, Meta has been forced to shift its focus to the rapidly evolving world of AI.
What sets Llama 3 apart from its competitors is its accessibility. By making the model open-source and leveraging its vast user base across its platforms, Meta has the potential to bring AI to the masses at the click of a button. This approach is a stark contrast to the closed systems of its competitors, potentially leading to a thriving ecosystem of AI-powered tools and services.
Llama 3 boasts an impressive pedigree, having been trained on over 15 trillion tokens – a dataset seven times larger than that used to train its predecessor. Meta also employed reinforcement learning with human feedback (RLHF) to filter out racist and horrendous content, ensuring a more ethical and unbiased output.
It is important to acknowledge that while Llama 3 is a step forward for Meta, it is not without its limitations. The model is not yet multi-modal, and Meta will need to ensure that it can keep pace with the rapid advancements being made by its competitors. Additionally, the ethical implications of open-source AI models must be carefully considered.