Meta Platforms is gearing up to launch the latest iteration of its large language model, Llama 3, in July 2024. This AI system aims to address a significant challenge faced by its predecessor, Llama 2, which currently powers chatbots on various Meta platforms. Llama 2 often refrains from answering questions deemed controversial, even if they are harmless in nature.
To overcome this limitation, Meta researchers are striving to “loosen up” Llama 3. This means the model will be trained to provide context for queries it identifies as potentially contentious, rather than simply refusing to answer them altogether. This approach is expected to enhance the usefulness and versatility of Llama 3, particularly when navigating complex or sensitive topics.
The move comes amidst growing competition in the AI language model space, with Google recently pausing the image generation feature of its model, Gemini, due to concerns about historical image inaccuracies. Meta’s efforts to refine Llama 3 highlight the ongoing quest for more nuanced and informative AI systems, capable of handling a broader range of queries and user interactions.
Furthermore, Meta plans to appoint a dedicated individual within the company to oversee the model’s tone and safety training. This signifies Meta’s commitment to responsible AI development, aiming to ensure that Llama 3 operates within ethical and unbiased parameters.
While the official launch is slated for July, the development of Llama 3 and its broader implications for the AI landscape remains an ongoing process. It will be interesting to observe how this new model navigates the complexities of human language and user interactions in the real world.
