Demystifying the LLM

LLMs. What are they? How do they work? Whether you are new to AI, investigating to keep up with news and terms, or simply wanting a refresh, understanding LLM’s gives us particular context for the current state of AI.

What are they?

First, we must expound LLMs. LLM stands for “Large Language Model”. At the core, LLMs are advanced AI systems capable of understanding, generating, and interacting with human language.

What makes them large? Large refers primarily to the vast amount of data they are trained on. They learn from a diverse range of text sources such as books, articles, websites and social media, and this enables them to grasp the nuances of language. Imagine talking to someone who understood all the world languages, who understood context, and was able to retrieve information with near instant recall.

Language here refers to human language. All of it. All language and sources that are made available have become training data for these Models.

How do LLMs work?

LLMs are a type of neural network known as a Transformer. A neural network’s inspiration is the human brain.

So if you think about your brain as you have developed over your lifetime you have been able to recognize patterns, interpret data, learn from it, and a host of other daily functions.

Now, you are one person. Your life is built upon interactions, circumstances, situational interpretations, information gathering, wisdom from those who’ve lived longer and so on. As a humans, we do not have the capability or capacity to handle everyone’s thoughts and use of language.

As an example, imagine walking into your local library saying, “I want to check out your whole library.” That is not only impossible but also shows the level to which the information in the world is vastly beyond one person’s ability to digest it.

Our Digital Age

Great efforts have been made to digitize the majority of the worlds information. Libraries now have online pdf’s of books that at one point had to be sifted through manually.

And machine’s are built to handle data at a scale unimaginable to our minds.

And libraries are now just one sources. Imagine one social media’s archived history of conversation. How much conversation happens in a single day on social media that is now available data to be digested by Model’s and used for context building?

Hint: A lot

Attention Mechanisms

The concept of Attention Mechanisms behind LLMs is vast. For our purposes we will imagine it like indexes in a book.

Along time ago in a….if you wanted to look up information quickly, you would have to check out or buy a book. In many cases we would read through the entire book looking for the information we wanted. This inevitably led to too much information and trying to distill the whole book.

That is fine if the book is interesting and you’re reading for leisure. But often we are wanting quick information and feedback. We have all had some level of the following conversation:

You: “Hey what was this actor’s character name in this movie?” Friend: “I don’t know, you could probably just Google it.”

There was a point when Google was not available and we probably just kept a mild level of frustration until we figured out the information we wanted either through remembering or some other way. Everything has changed. Not only are our parents telling us to Google things now instead of dialoguing, LLMs have indexed the world’s data leading to virtually instant answers.

Through attention mechanisms LLMs are able to focus on parts of the text to get contextual understanding then generate coherent and relevant responses. This is similar to having a conversation with someone and picking up on inflection or tone and discerning a “true meaning” behind a statement.

For a much deeper technical dive on Attention Mechanisms please refer to this paper Attention Is All You Need

Applications of LLMs

Most likely you have already experienced many applications of LLM’s either through using ChatGPT, having a company Chat Bot built off company data, a virtual assistant for content creation, or a virtual assistant for academic research.

Workflows: Your Own Jarvis, ie Scout

These applications of LLM’s are quickly making them invaluable tools for daily workflows. With these assistants and agents becoming the “doer’s” we are able to become the thinker’s and director’s.

Optimizing workflows allows you, dear reader, to focus on adding value to your day, business, life. How does AI fit into your daily workflows? If we look at what AI excels at we find that AI is great at offloading cognitive overhead. We need to move past the notion that using AI will make us dumb or lazy and instead adopt the mindset that AI can help us learn what we need to learn and offload the stuff of life we don’t need to think about to a more capable third party.

Think of Jarvis in the Iron Man movies. Tony Stark needed Jarvis but without Tony Stark, Jarvis had no direction. Neither could have succeeded without the other.

Many of us may have struggled to have good workflows. Why? Typically our days have been more “work” than “flow”. A workflow indicates something similar to “in the groove” or “in the zone.”

Scout can do the work while you stay in flow. “Yeah, but what does that mean?” I hear you.

Ask yourself these questions:

What are parts of your business that are necessary but you dread doing? Do you spend a lot of time answering the same questions over and over? Tech company, do people ignore your documentation and go straight to your community channel and ask questions that are clearly answered in the documentation if they would just….

Ahh, zen and peace and good feelings to all if that would all be take care of.

Well, good news. Scout has arrived!

Conclusion

Large Language Models are transforming the way we interact with technology, offering unimaginable possibilities and innovative ways of engaging with information. As they continue to evolve, they hold the promise of making our interactions with machines more natural and intuitive.

Scout is leveraging LLM’s for “the rest of us” making it simple to set up an agent, create a page crawler, and a host of other offerings. Scout offers the ease of a “no code” tool with full customization through code should you so choose.