Meta has unveiled its new AI large language model, Llama 2, which can be used freely for both commercial and research purposes and operates on a similar level to ChatGPT.
Llama 2 has been released as multiple models, with a 7-billion (7B), 13-billion (13B), and 70-billion (70B) parameter sizes and fine-tuned chat variants of each.
Developers have already made Llama 2 available via API, and the model is likely to become even more widely used as it is embraced by businesses.
In benchmarks comparing the model to other open-source models, Llama 2 performed at a consistently higher level particularly when it came to math and benchmarks that probed the depth of its knowledge base.
Meta has also compared Llama 2 70B to GPT-3.5, which is used as ChatGPT’s primary model, and achieved similar results across all areas except for math problems. It also almost completely outperformed Google’s PaLM but was eclipsed by GPT-4 and PaLM 2.
Llama 2 70B particularly excelled at Multi-task Language Understanding, a ‘zero-shot’ benchmark intended to stress test a model’s pre-training knowledge by presenting it with tasks that fall outside of the data on which it was trained.
Despite being less powerful than ChatGPT in some instances, it may prove more beneficial to companies looking to adopt AI at zero cost, although applicants whose products and services have more than 700 million monthly active users must approach Meta directly for permission to use the model.
The policy has drawn the ire of some commentators, who have argued that it limits access by certain competitors.
Ah yes, the well known "except you, FAANG" clause that's so common in *open source* licenses like GPL, MIT, BSD, Apache2, ...Here I go again, this can't be for real lol https://t.co/XnuKhpxV6j pic.twitter.com/I7OfhCIj4nJuly 18, 2023
However, a number of high-ranking individuals from companies and institutions throughout the AI ecosystem signed a statement in support of Meta’s approach to AI:
“We support an open innovation approach to AI,” read the statement.
“Responsible and open innovation gives us all a stake in the AI development process, bringing visibility, scrutiny, and trust to these technologies. Opening today’s Llama models will let everyone benefit from this technology.”
The open technology non-profit group OpenUK has also registered its support for Llama 2, but acknowledged that the access limitations on the model prevent it from being fully classified as open source.
RELATED RESOURCE
Automate personalization with AWS
Learn how to leverage the benefits of personalization and automation across every touch point.
“Supporting this step by Meta to formally open up an LLM, offering open development and open innovation in this space, is a no-brainer for OpenUK and had unanimous support from our Board,” said Amanda Brock, CEO at OpenUK.
“The democratization of AI and opening innovation around it to enable collaboration across our open communities is an essential step in the future of this most impactful of technologies. Whilst we recognize that this license is not an official ‘open source software’ one, it provides certainty for those that want to take advantage of Llama 2. This is a fantastic step in the right direction for open AI communities.”
Llama 2 was trained between January and July 2023, and Meta stated it used an offline dataset. It also trained a 34-billion parameter-size model for testing, which features in its Llama 2 academic paper, but will not be released.
Meta has committed to releasing further versions of tuned Llama 2 models “as [it improves] model safety with community feedback”
The model is being distributed by Microsoft as part of an existing partnership with Meta, and will be available via Azure AI and optimized to run on Windows systems.
It will also be made available through the platforms of AWS and Hugging Face to suit a wide range of use cases throughout the market and open-source community.
At the time of launch, Meta listed partners and supporters including Accenture, Jio, IBM, DropBox, Nvidia, Intel, and AMD.
Qualcomm and Meta also announced that they will collaborate on bringing Llama 2 to mobile devices in 2024, to run AI workloads directly on devices for a localized, private experience.
Llama 2’s precursor LLaMA was also open source, available on request and intended primarily for research purposes.
A week after it was announced, the model was leaked on 4chan and consequently widely shared, with some calling for it to be stored on Bitcoin for perpetual access.
Researchers had warned that the leaked model could be used for malicious purposes, such as social engineering.
-
Meta executive denies hyping up Llama 4 benchmark scores – but what can users expect from the new models?News A senior figure at Meta has denied claims that the tech giant boosted performance metrics for its new Llama 4 AI model range following rumors online.
-
The DeepSeek bombshell has been a wakeup call for US tech giantsOpinion Ross Kelly argues that the recent DeepSeek AI model launches will prompt a rethink on AI development among US tech giants.
-
Meta won't release multimodal AI models in Europe due to "unpredictable" privacy regulationsNews The "unpredictable nature" of EU regulations have prompted Meta to scrap plans to release its models across the region
-
Meta's Llama 3 will force OpenAI and other AI giants to up their gameNews The new model pushes open source as a serious contender in the AI space, and proprietary models might soon find themselves playing catch-up
-
A new LLM jailbreaking technique could let users exploit AI models to detail how to make weapons and explosives — and Claude, Llama, and GPT are all at riskNews LLM jailbreaking techniques have become a major worry for researchers amid concerns that models could be used by threat actors to access harmful information
-
Three open source large language models you can use todayNews Enterprises are flocking to open source large language models, many of which have become highly popular - here’s three you might want to try out
-
An open source challenger to GitHub Copilot? StarCoder2, a code generation tool backed by Nvidia, Hugging Face, and ServiceNow, is free to use and offers support for over 600 programming languagesNews StarCoder2 offers code generation support for over 600 programming languages, and it’s free to use
-
Google’s new ‘Gemma’ AI models show that bigger isn’t always betterNews Smaller AI models are clearly the hot new commodity as Google unveils two new lightweight models, Gemma 2B and Gemma 7B
