Meta Releases LLaMA 2, to Keep AI "Competitive"

[The image above is generated by Midjourney. The prompt I used to create the image is listed at the end of this email.]

Interestingly, last week’s edition of The Artificially Intelligent Enterprise was focused on Fine-Tuning LLMs for specific use cases and enterprises. Now, the headlines in AI are around Meta’s newest release of their foundation model, LLaMA 2. Here’s the scoop.

Meta, formerly Facebook, has recently declared its plan to open up the code for its newest AI technology, LLaMa 2, to developers worldwide. However, there's a twist: companies with a monthly active user base of 700 million or more must secure a license from Meta to utilize the technology. This condition appears to be a strategic move to prevent major tech rivals from capitalizing on Meta's AI. Snapchat, TikTok, and WeChat notably surpass this threshold. So while it provides a free powerful tool to most, it potentially keeps competitors at bay.

This move is not unexpected, given Meta's history of competition and the ongoing scramble to acquire AI technology in the industry. Despite the proclamation of open-source, Meta has inserted a clause in the LLaMa 2 terms that limits its use by competitors. This has ignited discussions about the true meaning of "open source" software, with some arguing that any restrictions on software usage directly contradict the open-source concept.

Meta's choice to open source its technology is a testament to its resolve to stay competitive in AI. Other tech behemoths, such as OpenAI, have kept their advancements under wraps due to safety considerations and to ward off competition. Meta's "open" model seems to have factored in competition.

Nick Clegg, Meta's President of Global Public Policy, has voiced his opposition to the monopolization of foundational technology by a select few corporate giants. He underscored the strategic benefits for companies that opt for the release of open-source software. Meta's decision to open source a capable model could challenge competitors like Google or OpenAI regarding pricing.

In the LLaMa 2 announcement, Meta disclosed partnerships with companies such as Spotify, LG, and Qualcomm. The licensing clause in the open source announcement appears to be a strategic maneuver to foster collaboration from current and future partners that are too large to either sue or acquire.

The release of LLaMa 2 under a more lenient license permitting commercial use is commendable. However, it's not an approved open-source license, and it comes with some intriguing restrictions, which seem to have anti-competitive implications. This license doesn't seem to be based on a standard Open Source Initiative (OSI) approved license. It has specific terms and conditions unique to the Llama 2 model and Meta. The license includes some restrictions, such as not using the LLaMa materials to improve any other large language model (excluding Llama 2 or derivative works thereof). This is likely to prevent companies like OpenAI from leveraging their work. You must request a license from Meta if you have more than 700 million monthly active users. This seems to be aimed at Amazon Web Services (AWS), Microsoft, and Google for now but puts controls in place for anyone that grows to challenge them. Meta retains the right to terminate the agreement if you breach any terms or conditions.

Tip of the Week: Hugging Face, Silly Name, Good AI Resources

With all the talk around Meta’s LlaMA release, many companies are considering using this powerful Large Language Model to create their own Enterprise ChatGPT.

Hugging Face is a leading platform that provides tools and resources for businesses to leverage artificial intelligence (AI), particularly in Natural Language Processing (NLP). As a business owner, you can use Hugging Face to evaluate AI models, which can help you make informed decisions about implementing AI in your business.

1. Understand the Basics: AI models are complex systems trained to perform specific tasks, such as understanding text (NLP), recognizing images (computer vision), or making predictions (machine learning). Hugging Face specializes in NLP, offering a library of pre-trained models to understand and generate human language.

2. Visit the Model Hub: Hugging Face's Model Hub is a vast repository of pre-trained models. Each model has a detailed description, performance metrics, and user reviews. As a business owner, you can use this information to evaluate which models might suit your needs.

3. Consider Your Business Needs: Are you looking to automate customer service with a chatbot? Or maybe you want to analyze customer sentiment from reviews? The right model will depend on your specific use case. For instance, GPT-3 is great for generating human-like text, while BERT is excellent for understanding the context of text.

4. Test Models Online: One of the great features of Hugging Face is the ability to test models directly on their website. You can input your text and see how the model responds. This can give you a firsthand understanding of the model’s capabilities.

5. Check the Metrics: Each model on Hugging Face comes with performance metrics. These can include accuracy, precision, recall, and F1 score. These metrics provide a quantitative measure of the model's performance, which can be crucial when comparing different models.

6. Community Reviews and Use-cases: Hugging Face has a large community of AI researchers and practitioners. Many users leave reviews about the models and share their use cases. These can provide valuable insights into each model's practical applications and limitations.

7. Integration and Implementation: Once you've chosen a model, the next step is to integrate it into your business operations. Hugging Face provides comprehensive documentation and code examples to help with this. However, depending on the complexity of your use case, you might need to hire a team of data scientists or AI engineers.

What I Read this Week

• Microsoft Reveals Guidance Ahead of Copilot Launch - UC Today

• Survey reveals mass concern over generative AI security risks - CSO

• The Last Word on AI and the Atom Bomb - Wired

• With Nvidia’s Help, Revenue Surges at Smaller Cloud Providers - The Information

• Long Sequence Modeling with XGen: A 7B LLM Trained on 8K Input Sequence Length - Salesforce AI Research

What I Listened to this Week

AI Tools I am Evaluating

• Maxme.ai - I have been using Maxme.ai since I found it on Product Hunt; it integrates your favorite chatbot into your web browsing experience - ChatGPT, Bard, Claude, and Bing AI.

• Microsoft Edge - I know this is strange, but using the Bing AI Chat allows me to create a comparison table from a chat, and the only way to use that is with Microsoft Edge. It’s also an excellent thrifty option for those who don’t want to pay for ChatGPT Plus.

• GPT-Trainer - Build your own AI assistant who will deliver the information you need right when needed. No coding required

Midjourney Prompt for Newsletter Header Image

For every issue of the Artificially Intelligent Enterprise, I include the MIdjourney prompt I used to create the header for the current edition.