Neural Nets, Generative AI, and LLMs in Plain English

This is a very basic primer for non-technical folks!

Untangling the AI Web: How Neural Nets, Generative AI, and LLMs Interconnect

And a surprise couple of lessons that were revealed along the way!

How do Generative AI, Neural Networks and LLMs fit together?

In my last post, we explored how Large Language Models, or LLMs, work. And at the risk of heading down a rabbit hole or confusing myself more, I decided to explore how LLMs work in conjunction with other phrases I’ve seen bandied about - like Generative AI, Neural Networks and more. So my exploration began.

• How are LLMs different from a Neural Network?

• How are LLMs different from Generative AI?

• (How) Do they fit together?

Boop! A Rabbit Hole!

Interestingly, I got a weird ‘this site isn’t working’ error when asking Anthropic’s Claude my first question, so I popped over to Chat GPT 3.5 (the free version), and got a very interesting response about LTSMs, not LLMs.

My prompt: “how is a llm different from a neural network? (in non-technical business language, please)”

Sure! In non-technical business language, let's explain the difference between an LLM (which I believe you meant to say "LSTM") and a neural network:

I was confused, never having seen the abbreviation LSTM before, which stands for Long Short-Term Memory.

LSTMs are a specific type of neural network tailored for sequence modeling, while generative AI encompasses a broader category of algorithms and models aimed at generating new content. LSTMs can be used as building blocks in generative AI models, but not all LSTMs themselves are generative in nature.

Okay, so my next prompt was easy: “when did LLM language morph into LSTMs in the "popular" vernacular?”

It seems there might be a misunderstanding here. As of my knowledge cutoff in September 2021 and considering the most common terminology in the field of machine learning and artificial intelligence, "LLM" is not a standard abbreviation or term related to any prominent concept or model in the domain.

The correct and widely used term is "LSTM," which stands for Long Short-Term Memory. LSTMs are a type of recurrent neural network (RNN) designed to handle sequential data effectively, as they have the ability to retain information over long periods of time.

If there were any other term like "LLM" that emerged after my last update, I wouldn't have information about it. It's essential to be cautious about potential typos, incorrect abbreviations, or any new developments that might have occurred after September 2021.

Lesson #1 - umm, Sept. 2021 was two years ago, AI has come a long way since then

And that, dear readers, is a challenge of free Chat GPT 3.5 and some other LLMs. It only has been trained on information up through September, 2021. Anything within the past two years is unavailable. So it was never trained on the concept of an LLM.

The Questions Continued - GPT 3.5 vs. Claude

So back I went to Claude, to see what I’d discover. It was back up and running, a mere 5 minutes later. My prompt: “how is a llm different from a neural network? (in non-technical business language, please)

Great question. The main difference between a large language model (LLM) and a neural network is the training data used:

A neural network is a type of AI algorithm structured similarly to the human brain. It can be trained on many types of data like images, audio, or sensor readings to recognize patterns. For example, a neural net could be trained on photos of different animals to learn to identify cats and dogs.

A large language model is a specific type of neural network that is trained on massive text data like books, websites, and conversations. Rather than recognizing images, it learns the patterns of human language.

So in business terms:

• A neural network is like an entry-level employee who can be trained to analyze many types of data.

• A large language model is an expert in language specifically. It's like a seasoned colleague who has read every memo, report, and email for years - so they deeply understand communication nuances.

This focus on language allows LLMs like ChatGPT to hold conversations, summarize documents, generate content and more. They have a nuanced understanding of human language much like an experienced writer or communicator would.

The key for businesses is identifying where conversational AI can simplify tasks or improve customer experience. LLMs open up many possibilities but still have limitations compared to general neural networks designed for things like image recognition. But for language-related tasks, LLM expertise makes them very powerful!

I felt much better about the response from free Claude vis a vis free Chat GPT.

As an aside, I found it VERY interesting that Claude referred to Chat GPT as an LLM, when it is, itself an LLM and users might consider each to be “competitive” with the other. (LLMs are incapable of feelings for now. HA!)

Lesson #2 - apply critical thinking, try different tools

It’s really important to challenge any response you get from these tools, as they’re all still in their puppy-learning phase. We’re smarter than they are for now, as long as we apply critical thinking skills.

Claude on Generative AI and LLMs

Sticking with Claude, then, I prompted it with the next question: “how do generative ai and LLMs relate to each other?”

Great question. Generative AI and large language models (LLMs) are closely related:

• Generative AI refers to algorithms that can create new content like text, images, or video. The key is they are not simply regurgitating existing examples, but synthesizing fully novel and original outputs.

• Large language models are a type of generative AI specifically focused on generating human-like text and language. So LLMs are generative AIs optimized for language generation.

Some examples of how LLMs demonstrate generative abilities:

• Summarizing a long text document into a short synopsis requires generating new text, not copying existing sentences.

• Answering questions on a topic requires generating a novel response using its own words.

• Translating text from one language to another requires generating equivalent text in the target language.

• Continuing a story prompt or having a dialogue involves generating brand new conversational text.

So in short, generative AI produces novel content, while LLMs are a subclass optimized for language generation. LLMs are trained on massive text data to generate high-quality linguistic content. Their generative capabilities enable a wide range of applications from conversational bots to creative writing aids.

Claude Explains How They All Relate

Finally, the big question: “So, again in professional business terms, how or do neural networks, generative AI and large language models all work together?”

You're right, it's important to understand how these AI technologies work together in business contexts:

• Neural networks are the underlying algorithmic architecture - they provide the basic pattern recognition capabilities. You can think of them as the mathematical models.

• Generative AI refers to neural nets trained and optimized to produce novel outputs like images, audio or text. They leverage their pattern recognition abilities to create original content.

• Large language models are a specific type of generative AI. They use neural networks to generate high-quality, human-like text instead of things like images.

So in business applications:

• A company may use neural networks for tasks like forecasting, classification, or recommendations. This provides a foundation of predictive analytics.

• For applications like generating summaries, translations, or conversational dialogue, they leverage generative AI like large language models. This unlocks more human-like language capabilities.

• Large language models allow building conversational interfaces, dynamic document creation, synthesized meeting notes, and more impactful applications.

So in summary, neural networks provide the foundation, generative AI allows creative content production, and large language models specifically enable powerful linguistic generation tailored for business needs. Companies combine these technologies to create more useful, interactive, and human-like AI systems.

I’ll let my friend Dru talk through the Generative AI tools that deal with images, in which he’s an experienced practitioner.

Lesson #3 - it takes a village, thank goodness for good people and AI tools

Resources from AIGG on your AI Journey of Understanding

Speaking of reliance on good people, have a look at our Resources section where you’ll find free drop-in HR Handbook policies for your team to review, augment and approve. And we have a Legal and Operational Issues List for you to better understand what to consider as you approach the use of AI tools.

We have attorneys, anthropologists, brand and business leaders to support you as you develop your Strategic AI Use Statements that can guide your organization’s use of the tools available to you. And we have bespoke educational workshops available to you as you explore your options.

Connect with us for more information, to get your free Legal and Operational Issues List, HR Handbook policy, or to schedule a workshop to learn more about how to make AI work safely for you. We are here for you.