Only a few months ago, we began seeing the tiny blue ring on our WhatsApp screens. And now, Meta is here with its new AI model. Named Llama 3.1, the open-source model implies that anyone can use and modify them without payment. The latest release includes a model trained on much more data than the previous version. Meta CEO Mark Zuckerberg said that Llama 3 is competitive with the most advanced models.
With the coming of ChatGPT back in 2022 big Tech companies like Microsoft and Google began relaying to develop new AI products. Meta also spent on AI. But instead of selling AI to other businesses, Meta decided to make its AI open source. This would help companies with no AI use Meta’s open model.
This is Meta’s largest text-based language model to date supporting eight languages and larger context windows. The new languages include French, German, Hindi, Italian, Portuguese and Spanish. The new model will be out on Meta and Hugging Face.
Rob Sherman, the vice president of policy and deputy chief privacy officer at Meta said that already Llama AI models have been downloaded by companies and individuals 300 million times. ‘Model distillation’ also makes the new model unique from proprietary closed vendors.
Everything to know about the Meta opensource AI model
As mentioned earlier, an open-source model helps anyone use and modify them without payment. It is publicly available. It can be used for commercial and non-commercial purposes. These models have open-source licenses.
Such models give the users the sole power to analyze how the system works and inspect the components. These models allow the developers to improvise and revise the recommendations which means the models undergo thorough revision catering to specific needs. These modified versions are later available to be shared with the public.
The open-source models come with datasets, ready-to-use interfaces and prebuilt algorithms to help developers embark on their app development journey. These models are also helpful for startups to increase their reach in the regional and global markets.
While the open-source model garners appreciation, it has also triggered discussions on safety and privacy concerns. Another challenge is the cost of deploying this model in large-scale enterprises. It is learned that Meta expects large-scale organizations to use smaller versions (70B and 8B hosted on the cloud).
Compared to Llama 3 models, Llama 3.1 is significantly more complex. It has 405 billion parameters and was trained with over 16,000 of Nvidia’s expensive H100 GPUs. The cost of developing the new model is still confidential. Based on the cost of the Nvidia chips alone, one can easily guess that it would cost hundreds of millions of dollars. Inspired by Linux, an open-source operating system, Zuckerberg states that open-source AI models will overtake the closed ones.
Ahmad Al-Dahle, Meta’s VP of generative AI, predicts that Llama 3.1 will be popular with developers as “a teacher for smaller models that are then deployed” in a “more cost-effective way.” If Meta’s new AI model becomes successful, it will undoubtedly undermine the business models of its Anthropic, Google and OpenAI models.
