Meta's new open-source large language model (LLM), Llama 3, is scheduled for release soon. It will include various models with different capabilities and versatility, with plans for a complete open-source launch in July to compete with other models like Claude 3 and GPT-4.
Meta's objective is to utilize Llama 3 to power a range of its products, aiming to create the most useful AI assistant. However, significant work remains to achieve this goal, according to Joelle Pineau, Vice President of AI Research.
Meta's Chief AI Scientist, Yann LeCun, advocates for the use of Joint Embedding Predictive Architecture (JEPA) over generative AI like Llama 3. JEPA offers improved training efficiency by discarding unpredictable information.
[
add
]
[
|
|
...
]