Meta, the renowned company, released the largest version of its AI models, "Llama 3," today. This new version is largely free, multilingual, and features general performance metrics that rival paid models from competitors such as OpenAI.
New Features and Performance
Meta, the owner of Facebook, announced in a blog post and a research paper that the new "Llama 3" model can speak eight languages, write high-quality computer code, and solve more complex mathematical problems than previous versions. The new release dwarfs last year's version, with 405 billion parameters or variables considered by the algorithms to generate responses to user queries. Although it remains smaller than leading models from competitors, it demonstrates substantial improvements.
Comparison with Competitors
For instance, OpenAI's GPT-4 model is said to have one trillion parameters, while Amazon is investing in a model with two trillion parameters. Despite its smaller size, the "Llama 3" model is engineered to deliver competitive performance.
Updated Models
In addition to the flagship "Llama 3" model, Meta Platforms announced updated versions of its two smaller models: one featuring eight billion parameters and the other with 70 billion parameters, introduced in the spring. These three new models are multilingual and can handle a larger number of user requests through an expanded "context window.
Improvements in Code Generation
Ahmad Dahla, Head of Generative AI at Meta, stated that these models will particularly improve the experience of generating computer code. He mentioned in an interview with Reuters that the extended context windows provide the models with something closer to long-term memory, which helps in processing multi-step requests.
Free Access and Strategic Implications
Meta largely allows developers to use the Llama models for free, a strategy that CEO Mark Zuckerberg believes will yield innovative products and increased engagement on the company’s core social networks. However, this has surprised some investors due to the associated costs. The company will benefit if developers choose to use its free models instead of paid ones, potentially limiting the use of competitors' models. Meta also highlighted the gains achieved in mathematics and core knowledge tests, which could make this option more attractive.
Market Impact and Future Outlook
While measuring progress in AI development is challenging, Meta's test results suggest that its largest models, Llama, are nearly on par with Anthropic’s Claude 3.5 Sonnet model and OpenAI’s GPT-4, and even outperform them in some cases. This release comes as tech companies race to demonstrate that their growing portfolio of resource-hungry large language models can achieve significant gains in challenging areas such as advanced reasoning to justify the substantial investments made in them.
AI
%20(703%20x%204_20241010_011310_0000.webp)
