① Llama 3 has two versions, 8B and 70B. The larger version of Llama 3 will have more than 400 billion parameters; ② More advanced reasoning skills, such as the ability to make longer multi-step plans, will appear in subsequent versions.
Financial Services, April 19 (Editor Niu Zhanlin) On Thursday local time, US tech giant Meta launched its most powerful open source artificial intelligence (AI) model, Llama 3, to catch up with industry leader OpenAI. In the US stock market, Meta shares have risen more than 2% and have risen nearly 43% so far this year.
Meta CEO Zuckerberg claims that Llama 3 has two versions, 8B and 70B, and that the larger version of Llama 3 will have more than 400 billion parameters. Llama 3 is a huge improvement over Llama 2 due to pre-training and fine-tuning instructions.
Llama 3 showcased cutting-edge performance on multiple industry benchmarks and provided new features including improved reasoning capabilities. Meta believes Llama 3 is the best open source big model on the market. Open source means that the code and data for these models are open to the public and can be viewed, modified, and used by anyone.
The developers complained that the previous Llama 2 model didn't understand the basic context and was often confused when processing queries. Google's Gemini AI image generation tool also experienced a similar problem, which generated inaccurate descriptions when generating images of historical figures, which drew widespread criticism.
Meta is now using higher quality data when training Llama 3, which helps the AI model to better recognize subtle differences in language, thereby improving its ability to understand the context.
Meta mentioned that the amount of data they entered for the Llama 3 model was seven times that of Llama 2, which may help improve the model's performance and accuracy. “Synthetic” data generated by AI is also used to enhance the model's ability in specific fields such as coding and inference.
According to Meta, Llama 3 will be integrated into its virtual assistant Meta AI, the most advanced AI application of its kind for free use. Meta AI Assistant has been launched in apps such as Facebook, Instagram, WhatsApp, and Messenger, and will be updated soon.
Meta's chief product officer Chris Cox said in an interview that the social media giant equipped Llama 3 with new computer coding capabilities. This time, in addition to being able to input text, it can also input images, but currently the model can only output text content. As a result, Llama 3 is currently not a large multi-modal model.
But he added that more advanced reasoning skills, such as the ability to make longer multi-step plans, will appear in subsequent versions. They are also planning to release multi-modal versions in the next few months, which means they can simultaneously generate text and images.
Cox said the ultimate goal is to help users get rid of complicated work and make life easier and happier, whether it's interacting with businesses, writing, or planning trips.
Additionally, Llama 3 will soon be launched on Amazon AWS, Databricks, Google Cloud, Hugging Face, Kaggle, IBM's cloud platforms WatsonX, Microsoft Cloud Azure, Nvidia's NIM and Snowflake, and will be supported by hardware platforms provided by AMD, AWS, Dell, Intel, and Nvidia.