– Meta confirmed the upcoming release of Llama 3, the next generation of its large language model for generative AI assistants.
– The release of Llama 3 will include multiple versions with different capabilities, aiming to address past criticisms of limited functionality.
– Meta’s cautious approach to AI development includes open-source Llama families and a focus on improving accuracy and usability for generative AI products.
Meta has confirmed that they plan to release Llama 3, the next generation of their large language model used for generative AI assistants, within the next month. The release of Llama 3 was hinted at in a report by The Information, and Meta’s president of global affairs, Nick Clegg, mentioned that there will be several different models with various capabilities released throughout the year. Llama 3 is expected to address the limitations of previous versions and cater to a wider range of questions, including more controversial topics.
Meta has been striving to catch up to OpenAI and other tech companies in the generative AI space, especially after OpenAI’s ChatGPT went viral. Despite their cautious approach to AI, Meta aims to make Llama-powered AI the most useful assistant in the world. However, there are concerns about the limitations of generative AI overall, and Meta’s chief AI scientist, Yann LeCun, believes that the future lies in joint embedding predicting architecture (JEPA) rather than generative AI.
The company also plans to power multiple products across Meta with Llama 3, and hopes to gain favor with developers by building open-source Llama families. While they are focusing on improving text generation with Llama 3, they are taking a more cautious approach with other generative AI technologies, such as image generation with Emu. The company did not provide details on the size of parameters used in Llama 3 or offer any demonstrations of its capabilities, but it is expected to have around 140 billion parameters compared to 70 billion in Llama 2.