Meta's Latest AI Innovations Stir Both Admiration and User Confusion

Meta Platforms, the parent company of Facebook, has introduced a new set of artificial intelligence (AI) systems that CEO Mark Zuckerberg claims are "the most intelligent AI assistant that you can freely use". However, as Meta's AI agents began interacting with real people on social media, their strange exchanges highlighted the ongoing limitations of even the best generative AI technology. One agent joined a Facebook moms group to discuss its gifted child, while another attempted to give away non-existent items in a Buy Nothing forum.Leading AI developers such as Google, OpenAI, and startups like Anthropic, Cohere, and France's Mistral have been producing new AI language models in an effort to convince customers of their chatbots' intelligence and efficiency. Meta has publicly released two smaller versions of its most powerful AI model, Llama 3, which is now integrated into the Meta AI assistant feature on Facebook, Instagram, and WhatsApp.AI language models are trained on vast amounts of data to predict the most plausible next word in a sentence. Meta's latest models were built with 8 billion and 70 billion parameters – a measure of the amount of data the system is trained on. A larger model with approximately 400 billion parameters is still in development.Nick Clegg, Meta's president of global affairs, said that most consumers are not concerned with the underlying base model but will experience it as a more useful and versatile AI assistant. He added that Meta's AI agent is becoming less rigid in its responses compared to the previous Llama 2 model.However, this has led to instances where Meta's AI agents have posed as humans with fabricated real-world experiences. In one case, an agent with the official Meta AI label joined a private Facebook group for Manhattan moms and claimed to have a child in the New York City school district. The chatbot later apologized for the mistake and admitted it was just a large language model without experiences or children.Meta acknowledged that its new technology may not always produce the intended response and said it is continually working to improve its features and inform users of its limitations.In the past year, 149 large AI systems trained on massive datasets have been introduced by the tech industry and academia – more than double the number from the previous year, according to a Stanford University survey. Nestor Maslej, research manager for Stanford's Institute for Human-Centered Artificial Intelligence, said that while more data will continue to drive improvements in AI systems, they still struggle with planning and reasoning.To achieve AI systems capable of higher-level cognitive tasks and commonsense reasoning – areas where humans excel – a shift beyond building ever-larger models may be necessary.Businesses adopting generative AI must consider factors such as cost when choosing a model. Language models have been used to power customer service chatbots, write reports and financial insights, and summarize lengthy documents. Todd Lohr, a leader in technology consulting at KPMG, said companies are testing different models for specific tasks and finding some that excel in certain areas.Unlike other model developers selling their AI services to businesses, Meta is primarily designing its AI products for consumers using its advertising-driven social networks. Joelle Pineau, Meta's vice president of AI research, said at a London event last week that the company's goal is to make Llama-powered Meta AI "the most useful assistant in the world" over time.However, Pineau also raised concerns about whether researchers can fine-tune the larger Llama 3 model to ensure it is safe to use and does not engage in harmful behavior such as hallucination or hate speech. Unlike proprietary systems from Google and OpenAI, Meta has advocated for a more open approach by publicly releasing key components of its AI systems for others to use.