.png)
December 8, 2024
Meta has released the latest version of its Large Language Model (LLM), Llama 3.3. The 70B model promises not only comparable performance to its predecessor, Llama 3.1 405B, but also significantly lower operating costs and improved efficiency. This development is particularly relevant for companies like Mindverse, which offer AI-powered content solutions and rely on powerful yet cost-efficient models.
Ahmad Al-Dahle, Vice President for Generative AI at Meta, presented Llama 3.3 on X (formerly Twitter) and published comparison data with other prominent AI models such as Amazon Nova Pro, Google Gemini Pro 1.5, and OpenAI ChatGPT-4o. The benchmarks included various tests, including "Instruction Following," "Long Context," and mathematical problem-solving tasks in multiple languages.
Llama 3.3 was particularly impressive in the area of "Instruction Following," meaning the precise execution of instructions. With a success rate of 92.1 percent in the IFEval benchmark, it shares the top spot with Amazon Nova Pro. In the "Long Context" area, which tests the processing of longer text inputs, Llama 3.3 also achieved a high hit rate of 97.5 percent. Compared to its predecessor, Llama 3.1, Llama 3.3 is marginally weaker in some areas, but this is compensated for by the increased efficiency and lower operating costs.
Llama 3.3 is freely available through platforms like Hugging Face and the Meta website. Meta is pursuing an open-source strategy with the Llama model family to promote the spread and application of AI technologies. So far, Meta has recorded over 650 million downloads of the Llama models. According to Meta CEO Mark Zuckerberg, around 600 million people use Llama every month. Certain conditions apply to commercial use, with platforms with over 700 million monthly active users requiring a separate license from Meta.
The open availability of AI models like Llama 3.3 presents both opportunities and challenges. It enables innovation and broad application, but also raises questions regarding responsible handling and potential misuse scenarios. For example, it was reported that the Chinese military is using Llama models. Meta responded by granting the US government access to the technology as well. In the EU, Meta is currently refraining from introducing a Llama version due to concerns about compliance with EU regulations. The development of Llama 4 is already underway and, according to Zuckerberg, will require a tenfold increase in computing power. Completion is expected in 2025.
The developments surrounding Llama 3.3 are of great importance for companies like Mindverse, which operate in the field of AI-supported content creation. Efficient and powerful LLMs are essential for the development of innovative solutions such as chatbots, voicebots, AI search engines, and knowledge databases. The improved efficiency and lower operating costs of Llama 3.3 could significantly facilitate the development and deployment of such applications and reduce costs. At the same time, the discussion about the regulation and responsible handling of AI models underscores the need for an ethical and transparent approach in the AI industry.
Bibliographie https://www.heise.de/news/Llama-3-3-Meta-veroeffentlicht-neues-KI-Modell-10191845.html https://stadt-bremerhaven.de/meta-llama-3-3-neues-ki-modell-verspricht-mehr-effizienz-bei-geringeren-kosten/ https://t3n.de/news/llama-3-3-meta-praesentiert-ki-modell-mit-spitzenleistung-zu-niedrigeren-kosten-1662188/ https://www.datacamp.com/de/blog/llama-3-3-70b https://the-decoder.de/meta-veroeffentlicht-llama-3-3-das-mit-weniger-rechenaufwand-mehr-leisten-soll/ https://www.finanznachrichten.de/nachrichten-2024-12/64021010-llama-3-3-meta-praesentiert-ki-modell-mit-spitzenleistung-zu-niedrigeren-kosten-397.htm https://www.handelsmeldungen.de/article/llama-3-3-meta-praesentiert-ki-modell-mit-spitzenleistung-zu-niedrigeren-kosten-01JEGFTE6GZT568MGBHRKZP0HS https://www.it-boltwise.de/metas-llama-3-3-70b-eine-neue-dimension-der-ki-entwicklung.html https://x.com/t3n/status/1865367303237341559