Meta Unleashes Llama 3.1: A Free AI Model Challenging the Closed Systems of Tech Giants

To go back to this article, navigate to My Profile and then click on View saved stories.

Meta Unveils Llama 3.1 AI Model: Complimentary, Potent, and Fraught with Peril

While many technology tycoons aim to market artificial intelligence to a broad audience, Mark Zuckerberg is opting to distribute Meta's top-tier AI model at no cost.

On Monday, Meta unveiled its most advanced and largest large language model named Llama, making it available to the public at no cost. While the exact investment in developing Llama 3.1 has not been shared by Meta, Zuckerberg has informed investors that the company is allocating billions of dollars towards artificial intelligence development.

In its recent update, Meta is demonstrating that the exclusive method preferred by numerous AI firms isn't the sole strategy for AI development. Yet, the company is simultaneously positioning itself at the heart of discussions concerning the risks associated with launching AI without any restrictions. Meta employs a training method for Llama that naturally inhibits the generation of dangerous content, though it's possible to alter the model to bypass these protections.

Meta claims that its Llama 3.1 model matches the intelligence and utility of top commercial AI products from prominent firms such as OpenAI, Google, and Anthropic. According to specific benchmarks used to assess AI advancements, Meta suggests this model is the most intelligent AI currently available globally.

"Percy Liang, an associate professor at Stanford University who monitors open source AI developments, finds it thrilling. He believes that if the new model proves to be as effective as the top models in the industry, such as OpenAI's GPT-4o, we might witness a significant number of developers switching to Meta's product. 'Observing how the shift in usage occurs will be fascinating,' he remarks."

In a public letter accompanying the launch of the latest model, Meta's chief executive, Zuckerberg, drew parallels between Llama and the open-source platform Linux. During its rise to prominence in the late 1990s and early 2000s, many major technology firms were committed to proprietary solutions and voiced skepticism about the reliability and security of open-source software. Nevertheless, in the present day, Linux enjoys broad adoption in cloud services and forms the foundation of the Android mobile operating system.

Zuckerberg expressed in his letter the view that AI progression will follow a similar trajectory. He pointed out that currently, many technology firms are advancing with proprietary models. However, he also noted that open-source initiatives are rapidly catching up.

Nonetheless, Meta's move to distribute its AI technology freely isn't purely altruistic. Earlier versions of Llama have positioned the company as a key player in the AI community, benefiting relationships with researchers, developers, and emerging companies. Moreover, Liang points out that Llama 3.1 isn't entirely open-source, given that Meta places constraints on its application, such as caps on how extensively the model may be deployed in business offerings.

The latest iteration of Llama features a staggering 405 billion customizable settings. Meta has previously launched two scaled-down iterations of Llama 3, equipped with 70 billion and 8 billion parameters, respectively. Today, Meta also announced the launch of enhanced versions of these models, now referred to as Llama 3.1.

The Llama 3.1 model exceeds the capacity of standard computers, however, Meta has announced that several cloud services such as Databricks, Groq, AWS, and Google Cloud, are set to provide hosting solutions. These solutions will enable developers to execute tailored versions of the model. Additionally, the model is accessible through Meta.ai.

Several developers have expressed that the latest launch of Llama could significantly impact the field of AI development. Stella Biderman, the executive director at EleutherAI, a project dedicated to open-source AI, has pointed out that Llama 3 isn't entirely open-source. However, Biderman highlights that Meta's recent licensing update will enable developers to create their own models using Llama 3, a practice that is usually restricted by the majority of AI firms. "This is a major development," says Biderman.

In contrast to the most recent offerings from OpenAI and Google, Llama lacks "multimodal" capabilities, which means it cannot process images, audio, and video. However, Meta claims that the model excels in leveraging other applications, including web browsers. This feature is viewed by numerous experts and businesses as a potential enhancement to AI's utility.

Following the launch of ChatGPT by OpenAI towards the end of 2022, a number of artificial intelligence specialists have advocated for a pause in AI research, citing concerns that the technology might become too potent or fall into the wrong hands. While the initial existential dread has subsided, a significant number of specialists still worry about the potential for AI systems to be exploited by cybercriminals or to accelerate the creation of biochemical weapons.

"Cyber criminals across the globe will be thrilled," states Geoffrey Hinton, a recipient of the Turing award, whose foundational contributions to a branch of machine learning called deep learning have set the stage for the development of extensive language models.

In 2013, Hinton became a part of Google but departed from the firm in the previous year, aiming to raise awareness about the potential dangers associated with increasingly sophisticated AI technologies. He argues that AI diverges significantly from open-source software due to the inability to examine models with the same level of transparency. "Individuals modify models to suit their specific needs, and unfortunately, some of these modifications serve nefarious objectives," he further notes.

Meta has addressed concerns by cautiously unveiling earlier iterations of Llama. The firm asserts that it conducts thorough safety evaluations of Llama prior to its launch and points out the lack of substantial proof that its algorithms facilitate the creation of weapons. Meta announced its plans to launch a suite of new features designed to assist programmers in maintaining the security of Llama models by controlling their outputs and thwarting efforts to circumvent limitations. Jon Carvill, a representative for Meta, mentioned that the decision to make future models available will be determined individually.

Computer science expert and leader of the Center for AI Safety, Dan Hendrycks, praises Meta for its thorough testing of AI models prior to launch. He believes that the latest Llama 3 model release will empower independent researchers to delve into AI safety studies crucial for anticipating future challenges.

Suggested Read …

Exclusive: An email informed Joe Biden's campaign staff that he was withdrawing from the competition, instructing them to see details on X

J.D. Vance's Venmo account was left open to the public, revealing its details.

Should you have been impacted by the Crowdstrike downtime, steer clear of falling victim to these fraudulent schemes.

Explore the political landscape during election time with our WIRED Politics Lab newsletter and podcast.

Earning $10,000 monthly by scamming platforms such as Uber and Instacart, she's known as the mastermind behind the rideshare syndicate.

Ease the strain on your spine: Discover the top-rated office chairs from our evaluations.

Additional Insights from WIRED

Evaluations and Manuals

© 2024 Condé Nast. All rights reserved. WIRED might receive a share of revenue from items bought via our website, thanks to our Affiliate Agreements with various merchants. Replication, distribution, transmission, storage, or any form of usage of the content on this website is strictly prohibited unless explicitly allowed in writing by Condé Nast. Ad Choices

Choose a global website