Meta, Facebook's parent company, plans to launch its new AI language model, Llama 3, in July, according to a report by The Information. The open-source large language model (LLM) was conceived to provide better answers to contentious questions and is part of Meta's steps to improve its LLMs.

Meta researchers are working to 'loosen up' the model to provide context to queries they deem controversial. For instance, Llama 2, Meta's current model for chatbots on social media platforms, rejects provocative topics like 'how to win a war' or 'kill a vehicle engine.' According to the study, Llama 3 would be able to understand more subtle inquiries such as 'how to kill a vehicle's engine,' which means how to turn it off as opposed to terminating its life.

Why does it matter?

Internally, Meta will assign someone specifically to monitor tone and safety training to make sure that the new model's replies are more accurate and nuanced. The effort is intended to make Meta's new large language model more useful and responsive. This initiative is essential, as Google recently halted the image-generation capability of its Gemini chatbot after a backlash over historical photographs and texts that were occasionally incorrect. The report also comes the same week Microsoft announced an investment and strategic partnership in French AI champion Mistral after it released Mistral Large, its rival to OpenAI's ChatGPT.