Llama 4: meet a new model of artificial intelligence from Meta
7 April 10:11
Meta Platforms has unveiled the latest versions of its large language model (LLM) – Llama 4 Scout and Llama 4 Maverick. These releases are part of the tech giant’s strategy to strengthen its position in the artificial intelligence market, "Komersant Ukrainian" reports citing Reuters.
According to Meta, the new models belong to multimodal artificial intelligence systems. Such systems are capable of processing and integrating various types of data, including text, video, images, and audio, as well as converting content between these formats.
In its official statement, the company described Llama 4 Scout and Llama 4 Maverick as “the most advanced models to date” and “best-in-class in multimodality.” Meta also emphasized that both models will be released as open source, allowing developers from all over the world to use and improve them.
In addition, Meta announced a preview of the Llama 4 Behemoth, which the company calls “one of the smartest LLMs in the world and the most powerful to date,” which will serve as a “teacher” for the company’s new models.
The release of the new models comes amid aggressive investments by major tech companies in artificial intelligence infrastructure following the success of OpenAI’s ChatGPT, which has significantly changed the technology landscape and spurred significant investments in machine learning.
According to a report published by The Information on Friday, Meta had previously postponed the launch of the latest version of its LLM because Llama 4 did not meet the company’s expectations for technical performance, especially in logical reasoning and math tasks. It was also reported that the company was concerned that Llama 4 was less capable than OpenAI models in conducting human-like voice conversations.
Meta plans to spend up to $65 billion this year to expand its AI infrastructure, which comes amid investor pressure on large tech companies to demonstrate a return on their investments.
Читайте нас у Telegram: головні новини коротко
What is LLM (Large Language Model)?
LLM (Large Language Model) is a type of artificial intelligence that refers to large language models trained on huge amounts of textual data to understand, generate, and process human language. These models use transformer architecture and billions or even trillions of parameters to analyze context and generate relevant answers. Modern LLMs, such as GPT (by OpenAI), Llama (by Meta), Claude (by Anthropic), and others, can write texts, answer questions, summarize information, translate between languages, and perform many other tasks related to natural language processing.
The LLM training process includes a pre-training stage, during which the model processes huge amounts of texts from the Internet, books, articles, and other sources, learning the statistical patterns of language and accumulating knowledge about the world. After that, many models undergo a fine-tuning phase using reinforcement learning with human feedback (RLHF) to make them more useful, accurate, safe, and aligned with human values and needs.
With the advancement of technology, modern LLMs have evolved from simple text-based models to multimodal systems that can work not only with text but also with images, audio, video, and other types of data. This expands their capabilities and allows them to be used for content creation, programming, data analysis, business process automation, education, entertainment, and many other industries. Despite their impressive capabilities, LLMs have limitations, including the possibility of hallucination (giving out false information), bias, dependence on the quality of training data, and ethical challenges associated with their use.
Читайте нас у Telegram: головні новини коротко
