Generative AI and LLM’s are transforming the landscape of what’s possible online. In the last five years, models such as OpenAI’s ChatGPT and Alphabet’s Gemini have helped to usher in rapid growth in an age of technological change – with an estimated 86% of workers already using AI at work.
Recently, a new AI model has been introduced by Meta Technologies, the parent company of social media platforms Facebook, Instagram, and WhatsApp. Powered by a version of Llama 3, it looks set to introduce new opportunities for businesses that use Meta’s range of social media platforms.
What does it mean for social media users? No matter whether you’ve recently completed a Graduate Certificate in Project Management, or have been using social media for a long time, Meta’s new AI model aspires to provide a set of new high-quality tools for business users to use online. Let’s explore this new platform, as well as the opportunities where business owners may be able to use it to enhance their work.
What’s a Llama?
AI platforms are well known for a few things – their immense capabilities, and their silly names. After all, when you hear the name Llama, or Gemini, the first thing you think of isn’t normally about technology – perhaps you consider an animal or a horoscope.
In reality, Llama is much more complex than how it’s labeled on the tin. Introduced in early 2023, Llama is an acronym for the term Large Language Model Meta AI, and it’s part of a family of large language models, also known as LLMs.
In very simple terms, an LLM is a form of AI that can be used to work with unstructured data, like text, and using advanced probability models, estimate what words are most likely to be used next. This can then be used on massive datasets, and refined with further tools, to essentially take an input (known as a prompt), run it through a massive model of data (known as a pre-trained model), and output a (hopefully) considered response.
Llama is a model that runs off an extensive set of training parameters – in total, while the final model uses approximately 70 billion parameters, the Llama 3 model was itself trained on a total of nearly 15 trillion tokens.
What’s fascinating is that while the Llama 3 model is incredibly powerful, Meta has also made it openly available for researchers and other interested parties to use. This presents opportunities for those who have the necessary hardware to potentially make local models based on the Llama models.
How Does Meta AI Work?
Meta AI works as a context-driven chat-based generative AI tool. That may sound like a mouthful to pronounce aloud, but in reality, the way users engage with it is fairly simple. By opening a conversation with Meta AI, you can use text-based prompts to receive recommendations and results.
For example, by asking a question like ‘Where are some great local coffee places’, the AI will ask for some further information. Are you looking for a takeaway coffee, or potentially somewhere to sit in with a hot coffee and a biscuit? These responses feed back into Meta AI’s contextual engine – where it can then use these results, in combination with data from Meta’s various platforms and some search engines like Microsoft’s Bing, to provide a detailed response.
This could be helpful for a salesperson who’s looking to find places to meet with local business owners. It’s a handy niche use case, but it’s important to recognize that Meta’s platform has the potential to do a lot more.
Say, for example, an employee has a presentation that they’re looking to refine for a presentation. Meta AI can help to craft and refine paragraphs of text – which may be handy for those who like to write extensively and need to focus on some smaller areas.
It can also be used to create content – for example, prompting Meta AI to draw a picture of a duck will result in an image generated by the platform. Overall, Meta AI can do a wide variety of tasks – from generating programming language samples to providing recommendations, it’s an incredibly powerful platform.
Understanding the Limitations of LLMs
It’s important to understand that while the underlying LLMs that power Meta AI are powerful tools, they also come with limitations. While it might be nifty to ask for recommendations from an AI, it’s important to be aware that LLM responses do come with a chance of producing errors.
For example, let’s consider that coffee place query from earlier. Meta AI was able to use the prompts to provide me with a list of five ‘local’ coffee shops that were highly regarded in search reviews. What was interesting is that the AI didn’t seem to understand the concept of distance – with its first recommendation being a local coffee shop that was more than ten kilometers away (and not very local, either).
When you consider how LLMs work, based on probabilities, it makes sense that the model would not understand the meaning of distance. For end users, however, it can be relatively easy to interpret platforms such as AI as some sort of perfect tool – which is a challenge.
It’s important to recognize that AI has limits. While it’s a powerful tool, it shouldn’t be used as a substitute for well-generated content. AI can be buggy, and it’s certainly not perfect.
As a public model, it’s also important to recognize that commercially sensitive data should not be input into the Meta AI model. With the allure of AI comes challenges for large businesses to maintain control of sensitive business information, as companies like Samsung Monitor have learned. Employees must be data literate and understand why sensitive data should not be shared in open models.
For those who can understand the limitations and opportunities, Meta’s new AI tool provides a powerful way for people to embed AI in their ways of working. As AI continues to evolve in the workplace, it’s exciting to imagine what will be possible as future LLM models become available.
