top of page
tech360.tv

Meta Unveils Largest Llama 3 AI Model, Highlighting Language and Math Advancements

Meta introduces the largest Llama 3 AI model with improved language and math capabilities. With 405 billion parameters, the model excels in multilingual communication and complicated issue solving. Meta seeks to trump companies with free Llama models by prioritising innovation and involvement.

Meta AI logo
Credit: REUTERS

In comparison to its predecessors, the most recent Llama 3 model excels at conversing in eight languages, producing superior computer code, and solving more complex mathematical problems. With an astounding 405 billion parameters, this edition outperforms last year's release while being smaller than competitors' top models.


In contrast, OpenAI's GPT-4 model apparently has one trillion parameters, while Amazon is preparing to introduce a model with two trillion parameters.


Mark Zuckerberg, Meta's CEO, expressed optimism that further Llama models will outperform proprietary competitors by next year. The Meta AI chatbot, powered by these models, is expected to become the most popular AI assistant by the end of the year, with millions of people currently using it.


As IT giants strive to demonstrate the capabilities of their resource-intensive huge language models, Meta's top AI scientist predicts that these models will run into reasoning restrictions, forcing the use of alternative AI systems for breakthroughs.


In addition to the flagship 405 billion parameter model, Meta is releasing revised versions of its lighter 8 billion and 70 billion parameter Llama 3 models, which debuted earlier this year. All three models are bilingual and can manage larger user requests with an extended "context window," which improves the coding experience.


Ahmad Al-Dahle, Meta's head of generative AI, emphasised the need of bigger context windows in enhancing model memory for complicated demands like multi-step tasks.


Furthermore, Al-Dahle mentioned improvements in the Llama 3 model's performance on tasks such as math problem solving by using AI to produce training data.


Meta provides its Llama models to developers for free, with the goal of driving innovation, reducing reliance on competitors, and increasing engagement on its social networks. While some investors have expressed concern about the accompanying expenses, Meta predicts gains from developers preferring its free models over costly ones.


Meta's test results show that its largest Llama 3 model performs similarly to or better than top frontier models like as Anthropic's Claude 3.5 Sonnet and OpenAI's GPT-4o in key arithmetic and knowledge exams.


For example, on the MATH benchmark for competition-level math word problems, Meta's model scored 73.8, outperforming Claude 3.5 Sonnet's 71.1 but lagging GPT-4o's 76.6. Similarly, in the MMLU benchmark covering several subjects, Meta's model scored 88.6, trailing only GPT-4o's 88.7.


Meta researchers hinted about planned "multimodal" versions of the models that will incorporate image, video, and speech capabilities into the basic Llama 3 text model, seeking to compete with rival multimodal models such as Google's Gemini 1.5 and Anthropic's Claude 3.5 Sonnet.

 
  • Meta introduces largest Llama 3 AI model with enhanced language and math capabilities

  • Model boasts 405 billion parameters, excelling in multilingual communication and complex problem-solving

  • Meta aims to outperform competitors with free Llama models, emphasizing innovation and engagement


Source: REUTERS

As technology advances and has a greater impact on our lives than ever before, being informed is the only way to keep up.  Through our product reviews and news articles, we want to be able to aid our readers in doing so. All of our reviews are carefully written, offer unique insights and critiques, and provide trustworthy recommendations. Our news stories are sourced from trustworthy sources, fact-checked by our team, and presented with the help of AI to make them easier to comprehend for our readers. If you notice any errors in our product reviews or news stories, please email us at editorial@tech360.tv.  Your input will be important in ensuring that our articles are accurate for all of our readers.

bottom of page