The Gist
- Llama 2 launch. Meta’s Llama 2 shifts AI marketing dynamics.
- Tech titans. Microsoft and Meta jointly develop Llama 2.
- Legality matters. AI legislation impacts open-source AI use.
For months marketers have heard news from Google and Open AI regarding new generative AI model features.
But another player, Meta looks ready to dominate the headlines through the rest of 2023. It announced that it will release its second version of its AI model Llama as an open-source large language model (LLM) framework available for commercial and research purposes.
Doing so offers marketers and developers a way to create AI-enhanced apps and software with their own development. The possibilities of unique customer experiences with AI tailored to the needs of brands and their media is endless.
Related Article: Marketers Reveal Insights on Generative AI’s Impact on Digital Customer Experience
What Is Available With Llama 2?
Llama 2 is a collection of pretrained and fine-tuned large language models that offers a variety of foundational models. LLMs power the AI products that we see now. For example, OpenAI built a chatbot, ChatGPT, from its LLM, GPT-3.
Llama 2 users can download a series of files containing starter code for the pre-trained and fine-tuned modeling. Model weights are also included, as well as a starter chatbot, Llama2 Chat, designed for specialized dialogue use cases.
The predecessor, Llama 1, had restricted availability, only issued to research organizations on request. In contrast, Meta has made Llama 2 available through a number of platforms, Amazon Web Services, Hugging Face and Azure AI.
Microsoft Partnership
Microsoft is a major partner with Meta on the development of Llama 2. This is not the first such partnership between the tech titans. Both were responsible for PyTorch, a popular machine learning framework, and React, a highly popular JavaScript framework.
The evolution of Llama has been a growing narrative within the even faster moving narrative on AI.
Back in February Meta claimed Llama 1 achieved faster performance while being a smaller model compared to ChatGPT. Those promising results, along with 100,000 requests for access to the Llama 1, encouraged Meta to go on and develop Llama 2. The Llama 2 models were fine tuned on 40% more training data than Llama 1. The data consist of 1.4 million tokens — raw text extracted from the words in input prompts.
Parameters
Parameters are a key significance in ML and AI models. Parameters are variables of input data. Models use parameters to analyze tokens then generate intricate responses to those prompts. The number of parameters influence model performance. Larger models can handle more complex relationships among the tokens and produce more nuanced output.
But training models with a large number of parameters consume energy and computing resources. So smaller size models offer efficiency opportunities for optimizing the models with as few parameters as possible. Llama 2 models have a capacity of 7 billion to 70 billion parameters, while ChatGPT contains 175 billion parameters. This positions Llama 2 as LLMs optimized for applications with a smaller physical footprint, like a smart device.
Developer Resources
To guide budding Llama 2 developers, Meta has included a number of resources. There is a responsible use guide that outlines safety best practices that reflect the latest generative AI topics. Meta also released a transparency schematic research paper that discloses fine-tuning and evaluation methods used for the model and identify known challenges and issues.
View all
Related Article: How to Pick the Right Flavor of Generative AI
The Advantages of Open-Source AI
Llama 2 being an open-source LLM has advantages. Open source LLMs allow researchers to study their parameters and output and identify better ways to operate models. There is debate on how to best evaluate and compare LLM models, striking at the heart of many cautions about AI.
Frameworks are also arising to ease programming work, such as LangChain, a generic interface that provides prompt template, foundation models, and other processes that can be integrated into the model with user tasks.
Other Open-Source Releases
Llama 2 is not the only open-source AI release. There are other small scale LLMs such as Falcon and StableLM, the model behind the AI image maker Stable Diffusion. But these LLMs do not have a major tech company that can quickly establish developer communities. Usually when a programming framework is launched from the major tech companies, a large number of developers gravitate toward learning and exploring the frameworks, discovering innovative use cases, and establishing an accessible body of technical support.
Currently OpenAI and Bard are in a race to court developers as much as they are to acquire the general public’s acceptance. OpenAI has a ChatGPT plugin ecosystem, where developers have created add-ins to extend usability, while Google has encouraged developers to build onto the Bard API.
Poised for Considerable Audience
Llama2 is poised to be a well-known LLM among the public and developers. The smaller parameter volume and development potential of Llama 2 will draw a considerable B2B executive audience eager to create AI-infused services, software and programmable devices. Many of the platforms offering Llama 2 are standard tech infrastructure that companies use for their programmatic customer experience operations. The reliability of those platforms combined with the exploration possibilities of Llama 2 should give managers exciting AI application projects while fine-tuning data using familiar cloud systems.
A Challenge From Apple?
A main challenge to Llama 2 may come from Apple if it decides to release an open-source LLM. In reporting the rumor of Apple’s AI development, Bloomberg News noted that Apple executives had not decided how a potential AI framework would be released to the public. Microsoft’s investment in Llama 2 is a certain pressure for an Apple response to AI.
In the meantime, managers interested in using open-source frameworks like Llama 2 should pay closer attention to AI legislation that defines acceptable AI usage. For example, the European Union passed the AI ACT, which requires the establishment of safeguards to prevent AI systems from generating illegal content. AI providers must also abide by transparency audits, so that content or media inputs used to train algorithms are identified. For example, companies are barred from scraping biometric data from social media and inserting it into a database. Adoption of open-source generative AI will force companies to assess their data training and usage decisions against legal and ethical risks.
Final Thoughts on Llama 2
As businesses race to capitalize on the promises of AI, marketers should see a rise in revised strategies for incorporating AI and machine learning applications into the websites, apps and software essential to customer experience. Llama 2 is poised to lead the rising tide that will lift all AI-influenced boats.
