Meta's Llama 3.1 opens AI to all: Rural docs get life-saving tools, basement trolls get deepfake election nightmares, maintainers get zero pay

The open-source AI movement has made significant strides, with models like Llama 3.1 and Mistral Large 2 rivalling industry leaders. However, this democratisation also raises concerns about safety and security. Meta's release of Llama 3.1, a 405-billion-parameter model, has sparked debate about the potential for misuse. The model's open-source nature allows for fine-tuning, which can be used for both beneficial and harmful purposes. According to a report by the National Telecommunications and Information Administration, the lack of safety protocols in open models poses a significant risk. The EU's AI Act, which came into force on August 1, 2024, aims to address these concerns by providing a framework for AI governance. Meanwhile, organisations like Hugging Face and the Apache Software Foundation are working to develop guidelines and standards for open-source AI development. As of 2024, Hugging Face hosts over 250,000 model cards, but the quality and completeness of these documents vary greatly.