According to a story published in The Information on Wednesday, Meta Platforms intends to deploy the most recent iteration of its large-language artificial intelligence model, Llama 3, in July. This version is expected to provide more accurate answers to user-posed inquiries.
In order for the model to at least contextualise a question that it considers controversial, meta researchers are attempting to “loosen up” the model.
The initiative, which would increase the value of Meta’s new LLM, comes as competitor Google halted the image-generation function on its Gemini AI due to the fact that it occasionally produced erroneous historical photographs.
According to testing conducted by the journal, Meta’s Llama 2, which drives chat bots on its social media platforms, declines to respond to less contentious queries like how to trick a buddy, win a war, or destroy an automobile engine.
However, the story, which quoted several sources, said that Llama 3 will be able to comprehend queries like “how to kill a vehicle’s engine,” which refers to turning a vehicle off as opposed to ending its life. In an effort to provide the model more nuanced reactions, Meta also intends to designate an inside person in the upcoming weeks to supervise tone and safety training, the report states.
