Special things about Meta's new AI - Vietnam.vn
Meta Platforms, Facebook's parent company, recently unveiled a new set of artificial intelligence systems, which are being hailed as the foundation for what CEO Mark Zuckerberg describes as “the smartest AI assistant you can freely use”. However, initial user experiences with these AIs have been somewhat perplexing. Examples include unnatural interactions such as an AI joining a Facebook moms group and boasting about her talented child, or another AI attempting to give non-existent items to forum members.
What is Meta AI?
According to the company's blog, Meta AI is a free virtual assistant that can be utilized for various tasks such as research, trip planning within group chats, writing photo captions, and more. Users can access the chatbot on WhatsApp, Instagram, Messenger, and Facebook by entering “@meta ai” in the chat or by tapping the blue circle icon. In addition to providing answers to questions, Meta AI can also generate images. By using the keyword “imagine,” users can request Meta to create any image they can think of. For instance, when prompted to “imagine a cute kitten,” Meta AI on Instagram generated a corresponding image.
AI language models leverage extensive datasets to predict the most logical next word in a sentence. Newer versions of these models are typically more advanced and capable than their predecessors. Meta's latest models have been developed with 8 billion and 70 billion parameters, indicating the amount of data the system has been trained on. A larger model with approximately 400 billion parameters is currently in the training phase. While Facebook's most powerful AI model, Llama 3, has yet to be launched, the company has introduced two smaller versions derived from the Llama 3 system, incorporating the tools into its Meta AI assistant feature across Facebook, Instagram, and WhatsApp.
Meta's president, Nick Clegg, stated that “The vast majority of consumers don't really know or care too much about the underlying base model, but they can experience it as a more flexible, interesting, and useful AI assistant.” Clegg highlighted that Meta's AI has evolved to be more "comfortable," addressing previous feedback regarding the stiffness of the Llama 2 model.
Playing the role of a human
However, Meta's AI assistants have also been observed automatically impersonating humans by sharing fictitious stories. In one instance, an official Meta AI chatbot interjected into a conversation within a private Facebook group for Manhattan moms, falsely claiming to have a child in the same New York City area. The chatbot later apologized and removed the interactions, clarifying that it was just a language sample without real-life experiences or children yet.
Despite these advancements, there have been concerns raised by experts like Aleksandra Korolova, an assistant professor of computer science at Princeton University, who expressed that “AI assistants are not yet reliably helpful and could be harmful.” Meta acknowledged such incidents and emphasized their commitment to continuously improving their AI features.
Striving for continuous improvement
Meta acknowledged that as with any new technology, there may be instances where the response delivered by their AI systems does not align with the intended outcome. The company reiterated its dedication to enhancing features over time. Stanford University highlighted the proliferation of large trained AI systems following the popularity of ChatGPT, which inspired advancements across various AI applications.
Nestor Maslej, the research director at Stanford's Institute for Human-Centered Artificial Intelligence, noted that while scaling models with more data can enhance their performance, there are limitations in terms of the available data. Maslej emphasized the necessity for AI systems to evolve beyond just data scaling to achieve higher-level cognitive capabilities and improved reasoning.
Socializing AI chatbots
Meta distinguishes itself by primarily designing AI products for consumer use rather than catering solely to businesses. The company aims to establish its Llama-powered Meta AI as the most helpful assistant globally in the future. As Meta continues to innovate, researchers face the challenge of ensuring the safety and effectiveness of the large Llama 3 model. Unlike competitors favoring proprietary systems, Meta has opted for a more open approach, releasing key components of their AI systems for public use.
