Top of Form
Meta is making big strides in artificial intelligence, with Mark Zuckerberg announcing a slew of upgrades for Meta’s large language model and AI assistant. In a recent Instagram post, Zuckerberg revealed that Meta is focusing on improving its AI capabilities, aiming for ‘multimodality, multi-linguality, and bigger context windows’.
One of the key upgrades is the introduction of Llama 3, Meta’s latest AI model. Llama 3 comes in two sizes, 8B and 70B parameters, and has been integrated into MetaAI, the company’s AI assistant. Additionally, Meta has also released an image generator that updates pictures in real time as users type prompts.
These upgrades are part of Meta’s efforts to compete in the generative AI market, currently dominated by OpenAI. With these advancements, Meta is positioning itself as a major player in the AI space, offering users a more intelligent and versatile AI assistant.
Meta AI is on a mission to redefine the future with its advanced artificial intelligence assistant
Meta is stepping up its game with its virtual assistant, positioning it as a vanguard AI that surpasses its competitors in reasoning, coding, and creative writing. This move puts it in direct competition with Google’s AI models and even the French startup Mistral AI. The upgraded Meta AI assistant will be integrated into Facebook, Instagram, WhatsApp, and Messenger, and it’s rumored to have its own independent website, putting it head-to-head with OpenAI’s ChatGPT.
During a discussion about Meta AI’s latest advancements, Zuckerberg shared with Patel that the company is introducing a new version called Llama-3. He emphasized that these models are not only open-source for developers but also serve as the backbone of Meta AI. “We believe that Meta AI is now the most intelligent AI assistant available for free,” Zuckerberg remarked. He also revealed that Meta AI will integrate with Google and Bing to provide real-time knowledge.
Exciting New Updates
Mark Zuckerberg recently shared exciting updates from Meta AI, highlighting how they’re making AI more accessible across their apps. He mentioned that on Facebook and Messenger, users will soon be able to use the search box to ask any question, making information more accessible.
Zuckerberg also discussed new creation features, like animation, which allows users to animate any image easily. What’s impressive is that these features generate high-quality images quickly, even updating them in real-time as you type.
Speaking about Llama 3, Meta AI’s latest AI model, Zuckerberg explained that they have been training three versions: an 8 billion parameter model, a 70 billion model, and a 405 billion dense model, which is still in training. He highlighted that their 8 billion model is almost as powerful as the largest version of Llama-2 previously released. Additionally, Meta AI is also releasing its 82MMLU, boasting leading scores in mathematics and reasoning.
Zuckerberg emphasized that pursuing artificial general intelligence (AGI) has been a significant focus for Meta for quite some time. He noted that Meta established the Fundamental AI Research (FAIR) group a decade ago, which has since led to numerous advancements benefiting all their products. Recent years have seen notable breakthroughs like ChatGPT and advanced image generation models. To capitalize on these advancements, Meta launched the Gen AI group, aiming to integrate these technologies into their products and develop foundational models to enhance their offerings.
Meta’s AI models and coding structures
In a surprising move away from the current trend, Meta AI’s approach to AI models stands out. According to Zuckerberg, coding wasn’t a top priority during the development of Llama 2. He noted that users weren’t likely to ask Meta AI coding-related questions on WhatsApp. However, he acknowledged the shift in the last 18 months, with coding becoming crucial in various domains. Zuckerberg highlighted that even if users don’t ask coding questions, training the models on coding improves their ability to provide thorough answers and reason across different domains.
When asked about the future of AI and its potential to replace programmers, Zuckerberg emphasized a gradual evolution. He noted that the advancements being made will unfold progressively over time. “I’m not sure that we’re replacing people as much as we’re giving people tools to do more stuff,” he remarked.
Regarding the risks associated with open-sourcing powerful AI models, Zuckerberg affirmed Meta’s pro-open-source stance. However, he clarified that not everything they develop will necessarily be released to the public. “I’m basically very inclined to think that open-sourcing is going to be good for the community and also good for us because we’ll benefit from the innovations,” he explained. Yet, he added a caveat: “If at some point, however, there’s some qualitative change in what the thing is capable of, and we feel like it’s not responsible to open source it, then we won’t.”
FAQs
Yes, Meta is very pro-open-source. However, they reserve the right to withhold certain developments if they deem it irresponsible to release them.
Meta’s AI, including the latest Llama 3, is poised to advance progressively over time, offering users increasingly powerful and intelligent assistance.
