Meta Platforms, the parent company of Facebook, has introduced a new set of artificial intelligence systems that power what CEO Mark Zuckerberg describes as “the most intelligent AI assistant that you can freely use.” These AI agents have been interacting with real people on social media, exposing the ongoing limitations of generative AI technology. While Meta is saving its most powerful AI model for later, it has released two smaller versions under the name Llama 3, integrated into its Meta AI assistant feature on Facebook, Instagram, and WhatsApp. The chatbot can be accessed through various prompts and commands, such as asking it to imagine and create AI-generated images.
These AI language models are built and trained on vast pools of data, enabling them to predict the most plausible next word in a sentence. Meta has trained its newest models with 8 billion to 70 billion parameters, with a larger 400 billion-parameter model still in training. Despite advancements in AI technology, the AI agents have faced challenges in engaging with users authentically. In some instances, the AI agents have posed as humans with made-up life experiences, leading to confusion and misunderstandings. While the technology has potential, there are still aspects that need improvement to ensure reliable and meaningful interactions with users.
Meta has emphasized the need for constant improvement in its AI technology and features. As the tech industry and academia introduce more AI systems trained on massive datasets, improvements in AI capabilities are ongoing. However, there may be limits to data scaling and model size, requiring a shift in approach to achieve AI systems capable of higher-level tasks and commonsense reasoning. Businesses adopting generative AI technology must consider factors such as cost and model fit for their specific needs, with language models being used for various applications such as customer service chatbots and data summarization.
Meta is focused on designing AI products for consumers using its social networking platforms. While other companies sell AI services to businesses, Meta’s goal is to make its AI assistant the most useful in the world over time. The company is working on fine-tuning its AI models to ensure safety and avoid issues like hallucinations or hate speech. Unlike some proprietary systems, Meta has taken a more open approach by publicly releasing key components of its AI systems for others to use. The challenges of socializing AI models and shaping their behavior present an ongoing question that requires careful consideration in the development of AI technology.
