Meta has made its latest move in the AI wars, upgrading its AI assistant, Meta AI, with a new model that it claims is the best in its class.
background: In December 2022, OpenAI surprised the world (or at least the internet) with the release of ChatGPT, an AI chatbot that can understand and generate conversational text thanks to an underlying large-scale language model (LLM).
Since then, it seems like every other technology company has developed their own LLM and is racing to release even more powerful AI built on top of it.
Metaverse: Facebook owner Meta is no exception.
In February 2023, the company announced and open sourced Llama. Codes are available for free on the internet. By July, it shared an upgraded version (Llama 2) that powers the ChatGPT-like assistant called Meta AI, which was released in September.
“Meta AI will be the most intelligent assistant available for free.”
mark zuckerberg
Meta is upgrading Llama again, integrating new models into Meta AI and making it available to more people in more places.
“Llama 3 makes Meta AI the most intelligent assistant available for free,” Meta CEO Mark Zuckerberg told The New York Times. “And now that we’ve reached the quality level we want, we’re going to make this even more prominent and easier to use across all our apps.”
detail: Meta has open sourced two Llama 3 models. One has 8 billion parameters and the other has 70 billion parameters (generally, the more parameters, the more complex and capable the model is). According to Meta, in testing, it outperformed open source LLMs of comparable size.
“We believe these are the best open source models of their class at the time,” the company wrote in a blog post.
While developers use the new model to build their own tools and applications, non-programmers can now leverage Llama 3 through Meta AI to serve results from web search engines. This distinguishes it from the free version of ChatGPT. I can’t access the internet.
Meta AI can now be accessed through search boxes on Facebook, Instagram, WhatsApp, and Messenger, as well as on its own website and through Ray-Ban Meta smart glasses. According to Meta, the AI assistant will soon be integrated into the Meta Quest headset.
However, wherever you use Meta AI, the same issues that have always existed in LLM still apply. This means that AI can hallucinate and present false information as truth, so results should be fact-checked if necessary.
“We believe these are best-in-class open source models.”
meta
Future prospects: Mehta says he’s already training. 400 billion We plan to release new Llama models in the coming months, which are parametric versions of Llama 3 and include multimodal (capable of understanding and producing multiple types of media, such as text, images, and video) models.
Zuckerberg told The Verge that Llama 4 and 5 are also already in development. This represents Meta’s efforts to unseat OpenAI as the dominant player in his AI space.
“Our goal at this point is not to compete with the open source model,” he said. “Our goal is to compete with everything in the world and become the world’s leading AI.”
We look forward to hearing from you! If you have comments about this article or tips for future Freethink stories, please send an email to: tips@freethink.com.
