Meta’s AI Assistant Gets an LLM Update. Here’s What You Need to Know
Tech giant Meta this week released the latest generation of its large language model, Llama 3.1 405B. According to the company, it is an open-source model that is comparable to proprietary LLM competitors such as OpenAI’s GPT-4 and GPT-4o and Anthropic’s Claude 3.5 Sonnet.
Llama is what powers the Meta AI assistant. Starting Tuesday, Llama 3.1 405B will be accessible via the assistant found on Meta-owned WhatsApp and the Meta.ai site. While you can also use Meta AI on Instagram and Facebook, it wasn’t immediately clear whether the latest model would be available on those platforms. A spokesperson didn’t respond to a request for comment.
The first version of Meta’s Llama was released in February 2023, but even CEO Mark Zuckerberg admitted that earlier versions of Llama lagged behind their competitors.
“Last year Llama 2 was only comparable to an older generation of models behind the border,” he wrote in a blog post published on tuesday.
Large language models are the technology behind generative AI chatbots like OpenAI’s ChatGPT, Google’s Gemini, and Meta AI. They’re trained on huge datasets to learn how we use language, so they can generate their own unique content that sounds at least plausibly human.
In addition to the release of Llama 3.1 405B, Meta AI’s image generation feature, Imagine, will begin enabling what the company is calling “Imagine me” prompts, allowing you to create images of yourself doing things like surfing or as part of a surreal painting based on existing photos. Meta AI is also getting new editing tools, allowing you to remove and edit objects in images. Starting this week, English-speaking users will be able to share those images to Facebook, Instagram, Messenger, and WhatsApp.
One of the features that really impressed my CNET colleague Katelyn Chedraoui was Meta AI’s image generator, which she called “useful, but not impressive.”
According to Meta’s figures, Llama has been downloaded more than 300 million times to date.
Llama vs. Everyone
The latest Llama models, which also include Llama 3.1 8B and 70B, have a context window of 128,000 tokens, which is a measure of how much the model can remember in a given conversation. OpenAI’s GPT-4o and the recently announced GPT-4o Mini also have context windows of 128,000 tokens, while Google’s Gemini 1.5 Pro has a window of 1 million tokens.
According to a separate blog postLlama’s enhanced reasoning capabilities help Meta AI understand more complex questions — especially those involving math and coding. Meta models also support eight languages.
Like Llama 3, which was released in April, Llama 3.1 405B was trained on over 15 trillion tokens, which is equivalent to approximately 11.25 trillion words.
Meta says that the 8B and 70B models are best suited for textual summarization and as conversational agents and coding assistants. Meanwhile, 405B can be used to create synthetic data, or data generated by algorithms or via computer simulations (rather than coming from real-world sources). It can also be used in model distillation, which is the process of transferring knowledge from an LLM to a smaller model, offering AI capabilities and speed while taking up fewer computational resources.
More than 25 partners, including Amazon, Databricks and Nvidia, are launching related services for Llama 3.1 405B to support these developers. Zuckerberg believes this gives the model a better chance.
An important difference between Llama and its peers is that the Meta Model is open source. come in two variants. Proprietary LLMs can only be used by developers who purchase access. Open-source LLMs are freely available everywhere.
Zuckerberg said this will ultimately make Llama and Meta AI more competitive, much like how the open-source version of Linux software eventually became more popular than the closed, proprietary versions of Unix software developed by large tech companies. In his blog post, Zuckerberg argued that this is because Linux allowed developers to experiment and was more affordable, leading to more users and ultimately more progress.
