Mark Zuckerberg believes that Llama Meta AI could become the most widely used AI assistant. Meta has expanded its Llama model series with the introduction of Llama 3.3 70B, a new AI language model. This model is designed to match the performance of the previous top model, Llama 3.1 405B, but with significantly lower inference costs. Llama 3.3 70B is now available for download on the AI development platform Hugging Face and the official Llama website.
Meta reports that the Llama models are experiencing great success. Ahmad Al-Dahle, Vice President for Generative AI at Meta, shared a diagram on social media comparing the performance of Llama 3.3 70B with competing models. According to the chart, Llama 3.3 70B outperforms models like Google’s Gemini 1.5 Pro, OpenAI’s GPT-4o, and Amazon’s newly released Nova Pro in several industry benchmarks. This includes the Massive Multitask Language Understanding (MMLU), a benchmark that evaluates the language understanding of AI models and serves as an important performance indicator.
Llama 3.3 70B is Meta’s latest attempt to dominate the AI field with “open” models. However, the usage is only partially open, as Meta has set certain restrictions: platforms with more than 700 million monthly users require a special license to use Llama models. Despite these limitations, Llama is showing impressive numbers, with over 650 million downloads according to Meta.
Meta also uses Llama technology internally. The AI assistant Meta AI, which is entirely based on Llama models, reportedly has nearly 600 million monthly active users, as stated by CEO Mark Zuckerberg. Zuckerberg emphasizes that Meta AI is on track to become the world’s most used AI assistant.
Meta plans massive investments but leaves many questions unanswered. Despite the self-praise, Meta currently faces significant challenges. The company trains its AI models using public data from Instagram and Facebook users who have not explicitly objected to the use of their personal information. However, in the EU, these data are subject to GDPR regulations. Earlier this year, the EU asked Meta to temporarily halt the use of European user data for AI training until it could be verified whether the company complies with GDPR guidelines. Additionally, Meta itself expressed doubts about its ability to meet the requirements of the new AI Act, describing the implementation of the law as unpredictable and potentially problematic for its strategy of making Llama models openly available.
Besides legal issues, the financial demands are also enormous. To ensure the development and operation of future Llama models, Meta has announced the construction of an AI data center in Louisiana, costing 10 billion dollars. In the second quarter of 2024, Meta’s capital expenditures increased by almost 33 percent to 8.5 billion US dollars compared to the previous year, due to the expansion of server capacities, data centers, and network infrastructure. Whether this high expenditure will pay off remains to be seen. At least on the legal side, many questions remain open.