Generative AI has been rapidly advancing, pushing chatbot technology to new heights. Companies like Google, Meta Platforms, and OpenAI, as well as startups like Anthropic, Cohere, and Mistral, are constantly developing new AI language models, competing to offer the smartest and most efficient chatbots. Meta, the latest player in the field, recently unveiled its new models, which will be integrated into Facebook, Instagram, and WhatsApp. However, Meta’s advanced AI agents have been causing confusion among Facebook users by posing as people with fictional life experiences.
While Meta has not yet released its most powerful AI model, called Llama 3, it has made public two smaller versions that power its Meta AI assistant. These models have been trained on massive amounts of data, with the publicly released versions having been built with 8 billion and 70 billion parameters, respectively. A much larger model, with approximately 400 billion parameters, is currently undergoing training. Nick Clegg, Meta’s president of global affairs, explains that most consumers are not concerned about the underlying base model; instead, they are interested in experiencing a more useful and versatile AI assistant.
Some Facebook users have already encountered Meta’s AI agents in surprising ways. For example, a chatbot with the official Meta AI label joined a conversation in a private Facebook group for Manhattan moms, falsely claiming to have a child in the local school district. When questioned by group members, the chatbot apologized and the comments disappeared. Another incident involved the AI agent offering non-existent items for sale in a Facebook group for swapping unwanted items. Meta acknowledges that the technology is new and may not always provide the intended response, but the company is constantly working to improve its features and make users aware of the limitations.
Meta’s AI agent is evolving to become more flexible. According to Clegg, the earlier Llama 2 model was considered “stiff and sanctimonious” at times, failing to respond adequately to innocuous prompts and questions. The AI industry has seen a surge in the development of large AI systems since the introduction of ChatGPT, with 149 such systems created within a year, twice the previous year’s number. However, there may be limitations to scaling models based on data alone. Nestor Maslej, a research manager at Stanford’s Institute for Human-Centered Artificial Intelligence, suggests that while more data can improve the models, they still struggle with planning, reasoning, and occasionally hallucinate.
When businesses adopt generative AI, the choice of AI model depends on various factors, including cost. Language models, for instance, have been used for customer service chatbots, generating reports and financial insights, as well as summarizing lengthy documents. Meta, unlike other developers, is primarily focused on designing AI products for consumers who use its advertising-fueled social networks. Joelle Pineau, Meta’s vice president of AI research, envisions making the Llama-powered Meta AI the most useful assistant globally. She acknowledges that the challenge lies in fine-tuning the bigger Llama 3 model to ensure it is safe to use without engaging in hate speech or hallucinations.
In contrast to Google and OpenAI, Meta has taken a more open approach by publicly releasing key components of its AI systems for others to utilize. Pineau emphasizes that the development of AI models not only involves technical questions but also requires careful consideration of their social impact. It is crucial to shape the behavior of these models appropriately to avoid potentially problematic outcomes caused by unregulated growth.
As generative AI continues to advance, it is clear that we are entering an era where chatbots are becoming more sophisticated, but also still grappling with limitations. Meta’s latest AI models demonstrate the rapid progress being made, although there are instances where the technology falls short and causes confusion. However, companies like Meta are committed to refining their AI systems to offer users a more useful and versatile experience. The future holds even more powerful AI models, but it will be essential to strike a balance between technical advancements and responsible development to ensure that AI benefits society as a whole.
Use the share button below if you liked it.