| Developers: | Meta Platforms |
| Date of the premiere of the system: | July 2024 |
| Branches: | Information Technology |
Content |
2025
Plans to increase investment in AI infrastructure to $65 billion a year
Meta (banned in the Russian Federation) plans to invest $60-65 billion in 2025 on AI infrastructure (twice as much as in 2024).
Construction of a 2 GW data center in Louisiana and an increase in GPU to 1.3 million units. Meta plans to significantly increase its teams working on AI, which emphasizes the chosen direction of the company's strategic development.
The introduction of Llama 4 will begin sequentially from weaker to stronger models in the first half of 2025. Development is underway on a huge cluster of over 100 thousand GPUs of the latest generation of H100. Training requires 10 times more computing resources than Llama 3. Training requires 1 GW of capacity (comparable to the energy consumption of the city for 700 thousand people).
Meta plans to integrate Autonomous Machine Intelligence (AMI) into Llama 4, which will allow the model not only to perform tasks, but also plan, evaluate and adapt in real time, reasoning will also be introduced, as in OpenAI o1, DeepSeek and Gemini 2.
Meta plans to monetize AI. Chatbots for customer service: rate from $50/month per basic package. Premium subscriptions for users: access to advanced assistant functions ($9.99/month).
Meta actively uses AI to improve ad targeting, user preference analysis, and content recommendation systems, increasing conversion. The AI-based advertising platform (Advantage +) will continue to evolve, offering even more accurate targeting and automation of advertising campaigns.
Meta AI (chat bot), has already surpassed 700 million monthly active users, and is expected to reach more than 1 billion users in 2025.
Introducing AI into VR/AR devices (Quest 3 and Ray-Ban Meta) to improve interaction (object recognition, voice control, automatic text translation, auto-prompts). Plans to sell 10 million devices by the end of 2025
Learning from stolen books
Meta (recognized in Russia as an extremist organization and banned) used materials from the pirate library Library Genesis (LibGen) to train its Llama artificial intelligence model with the personal approval of CEO Mark Zuckerberg. This became known in January 2025.
According to The Register, information about the use of pirated content was disclosed during the trial at the suit of writer Richard Cadrie and other authors against Meta Platforms. The plaintiffs allege their works were illegally used to train artificial intelligence models.
Court documents indicated that the company had internal discussions about the ethics of using data from LibGen. The LibGen Library provides free access to academic books, bestsellers, audiobooks, comics and magazines. Currently, the resource is under threat of closure due to a lawsuit by large publishers demanding compensation in the amount of $30 million.
The case file also claims that Meta purposefully removed information about copyright holders from the materials used. Presumably, this was done to hide data sources and prevent leaks of information about the illegal use of content.
Meta claims transparency about its actions and denies hiding the fact of using the LibGen database. The company also disputes the plaintiffs' claim that information about the pirate library's use is new material in the case.
The lawsuit against Meta is part of a series of similar lawsuits targeting developers of artificial intelligence systems. The plaintiffs also include comedy actress Sarah Silverman.
During the pre-trial collection of materials, documents were found describing the internal debate of Meta employees over access to LibGen and subsequent approval of the use of the resource received from Mark Zuckerberg.[1]
2024
Rising spending on AI infrastructure
Released Llama 3.2 version and open source multimodal models
In Q3 2024, the Llama 3.2 version and open source multimodal models were released. Meta is actively collaborating with businesses and government agencies to implement Llama, including the U.S. government.
The company began developing Llama 4, training models on a cluster exceeding 100,000 Nvidia H100, one of the largest in the industry. Small models of Llama 4 are expected to be ready in early 2025.
Why do I need open source Llama? Does Mark Zuckerberg do charity work? No, the point is creeping comprehensive expansion and trying to create an ecosystem around Llama the way Apple created an ecosystem around its OS.
By introducing Llama for free, Meta is trying to achieve a monopoly position in AI projects that will predetermine development for the next few decades in the way it has created monopoly and global social media dominance.
By tying business, consumers and government structures to Llama, Meta will in the future try to control value chains in AI not through direct subscription, like OpenAI, but through services in the ecosystem - IT infrastructure, information and software binding, and so on.
The release of Llama 3.1, the world's largest open source AI model
July 23, 2024 Meta (recognized as an extremist organization; activities in the Russian Federation are prohibited) announced the release of the world's largest open model of artificial intelligence - Llama 3.1. It has 405 billion parameters and is said to surpass GPT-4o and Anthropic Claude 3.5 Sonnet in some characteristics.
Llama 3.1, according to Meta, is much more complicated than the previously released Llama 3 AI models. The training of Llama 3.1 involved 16 thousand powerful Nvidia H100 graphics accelerators. As of the date of the announcement, Meta does not disclose the cost of developing Llama 3.1. Market participants say that, based only on the cost of chips, Nvidia we are talking about hundreds of millions. dollars
In addition to the version with 405 billion (405V) parameters, the Llama 3.1 family includes models with 8 billion (8B) and 70 billion (70B) parameters. They have a context window of up to 128 thousand tokens. Models support English, German, French, Italian, Portuguese, Hindi, Spanish and Thai.
The Llama 3.1 8B AI model is suitable for use in conditions of limited computing resources. It can be used to solve problems such as text summarization, classification, translation from one language to another. The Llama 3.1 version is 70B suitable for content creation, conversational AI, language understanding and enterprise applications. The model, as stated by Meta, does a good job of summarizing, classifying text, analyzing, language modeling, generating code, etc. The most powerful version of Llama 3.1 405B is focused on the most complex problems, including mathematical calculations, generation of long texts, multilingual translation, etc. In addition, this model can be used in advanced enterprise-level services.[2]
Notes
| The site content is translated by machine translation software powered by PROMT. The machine-translated articles are not always perfect and may contain errors in vocabulary, syntax or grammar. Read original article If you find inaccuracies or errors in the results of machine translation, please write to editor@tadviser.ru. We will make every effort to correct them as soon as possible. |