Topic 5 What are Large Language Models in Artificial Intelligence?

Topic 5: What are Large Language Models in Artificial Intelligence?

Welcome to our artificial intelligence jargon explainer series! If you've been hearing a lot about artificial intelligence (AI) lately and want to understand some of the terms being used in this space, but you're not a tech expert, you're in the right place. Today, we're talking about large language models.

Artificial intelligence is becoming really adept at creating text that resembles human conversation, thanks to a technology called large language models, or LLMs. At their core, LLMs are a sophisticated type of artificial neural network designed to understand and generate human language, and they've transformed how humans interact with technology.

LLMs are pre-trained on massive, hence the term "large," datasets of text from books, websites, articles—pretty much anything that's ever been written down that could be loaded into the system. The initial training is unsupervised, but then some post fine-tuning is done in a supervised format. Check out video 3 to learn more about supervised and unsupervised learning.

What makes LLMs unique is that they typically use something called the Transformer architecture, which we will cover in a different video. But in a nutshell, Transformers are able to process and understand all of the parts of a sentence at once, rather than processing each individual word in sequence. This gives them the power to understand human language by weighing up the contextual importance of different words and phrases.

For example, if I asked you to finish the sentence "That person is so kind, they have a heart of...," some AI models might end the sentence with something biological, like "they have a heart of blood" or "they have a heart of muscle." However, LLMs will have come across this very human expression millions of times before and understand that, in the context of a person being kind, the correct word is "gold."

This amazing ability to understand context in words makes LLMs really powerful for translation between different languages. Rather than sequentially translating each word in a sentence, they can translate in a way that reconfigures for the word order of the language it's translating to.

Let me give you an example with two languages that I speak: English and Cantonese. The English sentence "I ate a delicious apple this morning" when translated to Cantonese by LLMs would translate and factor in the correct word order, which if I spoke back to you in English would actually say "This morning, I ate one delicious apple."

Large language models are a type of generative AI that has been specifically built to generate new text-based content, with the most well-known being ChatGPT. LLMs can be used to create automated customer support chatbots, help us to write emails and blogs, and even assist in writing computer code.

One word of warning though: large language models don't know everything but will confidently sound like they do. This means their output shouldn't be fully trusted and it's why humans are still very much needed in the world to double-check for these hallucinations or errors. Check out my next video where we will deep dive into topic six, hallucinations.

Keywords

• Artificial Intelligence (AI)
• Large Language Models (LLMs)
• Neural Networks
• Transformer Architecture
• Human Language
• Contextual Understanding
• Translation
• ChatGPT
• Generative AI
• Hallucinations

FAQ

Q: What are large language models (LLMs)? A: LLMs are sophisticated neural networks designed to understand and generate human language, transforming how humans interact with technology.

Q: What is unique about the Transformer architecture used in LLMs? A: Transformers can process and understand all parts of a sentence at once, giving them the power to weigh the contextual importance of different words and phrases.

Q: How do LLMs handle language translation? A: Rather than translating each word sequentially, LLMs reconfigure the word order for the target language, resulting in more accurate and natural translations.

Q: Can LLMs be fully trusted? A: No, LLMs don't know everything and may produce confident but inaccurate outputs. Human oversight is still necessary to check for errors and hallucinations.

Q: What practical applications do LLMs have? A: LLMs can be used to create automated customer support chatbots, assist in writing emails and blogs, and even help write computer code.