Dearest Gentle Reader - this week I try to answer a proverbial question.

One that has haunted humans for centuries. And that will affect AI for many years to come.

By now you have heard about Large Language Models (LLM's), Small Language Models (SLM's). But what is the difference between them, why does it matter?

And since we're all working in our summer bodies, and probably starving, I will use a food analogy:

Large Language Models (LLMs): The AI world's equivalent of an all-you-can-eat buffet. They've been trained on massive amounts of data and can handle a wide range of tasks. Think Chat-GPT or Claude - they're the Swiss Army knives of the AI world. From creating a poem about our inexistent summer to debugging some code.

LLMs are great for complex tasks and can understand context like a pro, but as covered in a previous edition a big con is their energy consumption.

Small Language Models (SLMs):If LLMs are buffets, SLMs are your favorite food truck - specialized and efficient. They're trained on specific types of data for particular tasks. An example is an AI companion that's an expert at analyzing legal documents?

SLM's are great in their specialty area, and they're much lighter on computational resources.

‍

When do you choose one over the other?

SLMs shine when you need speed, efficiency, and specialized knowledge. They're perfect for edge devices or when you need quick, specific responses. LLMs, on the other hand, are your go-to for complex, varied tasks that require a broad understanding of language and context.

Here's a snackable example (no pun intended): Imagine you're building a customer service chatbot for a bank. An LLM like GPT-4 could handle a wide range of customer queries, from explaining mortgage rates to suggesting investment strategies. But if you're specifically focused on quickly categorizing customer complaints, an SLM trained on banking terminology and complaint types might be faster and more accurate.

‍

So does size really matter? Well... it depends. On what you want to do with it.

Figure out what you're trying to solve for first. The first step to make AI work for your business.