Meta AI is the ‘most intelligent, freely-available’ AI assistant, says Mark Zuckerberg

Meta, in its bid to steer ahead of its peers in the generative AI arms race, has introduced a slew of upgrades. CEO Mark Zuckerberg reveals the features of Llama 3 and the company’s vision.

Meta CEO Mark Zuckerberg delivers a speech, as the letters AI for artificial intelligence appear on screen, at the Meta Connect event at the company's headquarters in Menlo Park, California, US. (File Image: REUTERS/Carlos Barria)

It seems Meta is on its way to make significant strides in artificial intelligence. On Thursday, Mark Zuckerberg in an Instagram post announced that Meta’s large language model and AI assistant are getting a slew of upgrades. Along with new upgrades, in a separate interview, Zuckerberg told podcaster Dwarkesh Patel that Meta was looking forward to bringing ‘multimodality, multi-linguality, and bigger context windows’.

Along with early versions of Meta’s latest LLM, Llama 3, the company also released an image generator that updates pictures in real time while a user types prompts. This is seen as Meta’s bid to catch up on the generative AI market which is currently dominated by OpenAI. The first models of Llama 3 have been released in two sizes, 8B and 70B parameters, and have been integrated into MetaAI, the company’s artificial intelligence assistant.

Meta AI aims to soar

Meta is currently pitching its virtual assistant as the sophisticated AI that is ahead of its peers in areas like reasoning, coding, creative writing, rivalling models owned by Google, and even French AI startup Mistral AI. The latest version of Meta AI assistant will be a prominent feature in Facebook, Instagram, WhatsApp, and Messenger apps. It will also reportedly have a standalone website making it a direct competitor of OpenAI’s ChatGPT.

View this post on Instagram

A post shared by Mark Zuckerberg (@zuck)

When asked about the latest models from Meta AI, Zuckerberg told Patel that people are going to see a new version of Meta AI, Llama-3. He said that the models are released as open-source for the developer community and will also be powering Meta AI. “We think now that Meta AI is the most intelligent, freely-available AI assistant that people can use. We’re also integrating Google and Bing for real-time knowledge,” he said.

Also Read | Meta’s Oversight Board seeks public view on AI-generated images of Indian, US celebrities

Describing the key features of the new releases from Meta AI, Zuckerberg said that Meta is going to make it ‘a lot more prominent across its apps’. “At the top of Facebook and Messenger, you’ll be able to just use the search box right there to ask any question. There’s a bunch of new creation features that we added that I think are pretty cool and that I think people will enjoy. Animation is a good one. You can basically take any image and just animate it. It now generates high-quality images so quickly that it actually generates it as you’re typing and updates it in real time,” he told the host.

Divulging more details on Llama 3, Zuckerberg shared that his company has been training three versions, the 8B parameter model 70B model, and another 405B dense model which is still training. The CEO said that his company is also releasing its 82MMLU and has leading scores in maths and reasoning. “The 8 billion (model) is nearly as powerful as the biggest version of Llama-2 that we released.”

On working towards artificial general intelligence (AGI) and if it will become a key priority for Meta, Zuckerberg said that it’s been a big deal for a while. He said that Meta started FAIR (Fundamental AI Research) 10 years ago. “Over the last 10 years, it has created a lot of different things that have improved all of our products. There’s obviously a big change in the last few years with ChatGPT and the diffusion models around image creation coming out. At that point we started a second group, the gen AI group, with the goal of bringing that stuff into our products and building leading foundation models that would power all these different products,” he told the host.

Meta AI models and coding

At a time when AI models are increasingly being benchmarked on their coding capabilities, Meta AI models seem to be a departure from the norm. Zuckerberg revealed that when they were working on Llama 2 they didn’t prioritise coding because, according to him, people weren’t going to ask Meta AI a lot of coding-related questions in WhatsApp. He acknowledged that over the last 18 months coding has become an important aspect in many domains “Even if people aren’t asking coding questions, training the models on coding helps them become more rigorous in answering the question and helps them reason across a lot of different types of domains.”

Story continues below this ad

Also Read | How to chat with Meta AI on WhatsApp

When asked at what point an AI model will be powerful enough to replace a programmer, Zuckerberg said that all the ongoing developments will be progressive over time. “I’m not sure that we’re replacing people as much as we’re giving people tools to do more stuff.”

On the potential dangers of open-sourcing powerful AI models, Zuckerberg asserted that Meta is very pro-open-source. However, he added that he isn’t committed to releasing everything they test. “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. 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.”