Meta Plans To Launch Next Gen-AI Llama 3
Meta has announced the launch of its next open-source large language model Llama 3 (Large Language Model Meta AI) in the next month. The news was disclosed at the AI Day event in London on Tuesday. Llama 3 will power various generative AI assistants, and will also power other products included in Meta. During the event, Nick Clegg who is the president of Meta’s global affairs said “Within the next month, actually less, hopefully in a very short period of time, we hope to start rolling our new suite of next-generation foundation models, Llama 3, which will have many different models with different capabilities, different versatilities.”
Earlier in July 2023, Meta announced Llama 2 which was also the first version that was made accessible to the public. It is an open-source model. The new Llama 3 is expected to have fixed some drawbacks of the previous version along with the introduction of new functions and features. This version is said to be a much bigger and more capable model than the previous one, which comes with over 140 billion parameters, as compared to the 70 million for Llama 2. It is also said to handle several questions of different genres, including controversial topics, and will answer questions with more accuracy.
Joelle Pineau, vice president of AI Research said “Our goal over time is to make a Llama-powered Meta AI be the most useful assistant in the world.” He further added, “There’s quite a bit of work remaining to get there.”
The introduction of the new Llama 3 is Meta’s catchup plan to compete with OpenAI’s ChatGpt-4, which is known for its potential to answer questions based on images. The next two smaller models of Llama 3 which are set to launch in the upcoming week will not have multimodal capabilities, whereas the biggest version is expected to be multimodal, and capable of processing both texts and versions.
Also Read: Apple’s Game-Changing AI System to Beat OpenAI
Like Meta’s previous generation models, the next generation AI model Llama 3 is also said to be an open-source model, wherein developers can use the model without incurring any costs and can include it in their day-to-day solutions. But Meta’s text-to-image/video generation tool Emu has not been made public to the users yet.
Here’s what one may expect from Llama 3
- The new large language model may have a larger context window as compared to Llama 2. A context window refers to the capacity of how much an LLM can see.
- LlaMa 3 can utilize MoE in smaller models, enhancing inference time and reducing RAM requirements, requiring a powerful PC for optimal performance.
- All LLMs are trained on data that operates in English which makes it difficult for international users to converse in their native language. This was the case with Llama 2 as well, where any language outside English became difficult. This development is much anticipated in the upcoming Llama 3 version.
- Llama 3 would support other mediums other than just texts, which implies that it would be available in the multi-modality mode. It would process mediums such as text, image, video, audio, IMU motion, sensor, etc.
