AI Is Not A Magic Human…

When people refer to AI in 2025, they are usually talking about Generative AI, specifically text-based Large Language Models (LLMs).

They are magical, but not in the way most people think or understand.

Let’s say you had:

“Sarah, Mike, John, Emily, Anna, Tom, Rachel, James”

In traditional computer programming, if you wanted to do anything useful with this text, you would need to split the names into a structured format using functions like “explode”, “split”, or “regular expressions”.

That would produce an output like this:

{
“names”: [
“Sarah”,
“Mike”,
“John”,
“Emily”,
“Anna”,
“Tom”,
“Rachel”,
“James”
]
}

But that only works because the text is structured.

The names “Sarah, Mike, John, Emily, Anna, Tom, Rachel, James” are clearly separated by commas.

However, once data becomes unstructured, it becomes much harder for traditional programming techniques to extract useful information.

For example, the following text cannot be easily processed with conventional programming alone:

“Last weekend, I met Sarah at the park, where she brought her dog for a walk. Later, Mike and I had lunch at a cozy café downtown. John and Emily joined us afterward to share some hilarious travel stories about their recent trip to Italy. Anna and Tom also came but didn’t really do much, and finally, Rachel showed up with a board game that kept us entertained for hours.”

A traditional computer program would struggle to extract who did what, because there are no clear separators like commas or rows in a database.

However, Large Language Models can “understand” the sentiment and context in human language/text/data and then infer structure from seemingly unstructured text, converting it into a structured format like this:

{
“people”: [
{
“name”: “Sarah”,
“activity”: “brought her dog for a walk at the park”
},
{
“name”: “Mike”,
“activity”: “had lunch at a cozy café downtown”
},
{
“name”: “John”,
“activity”: “shared travel stories about a recent trip to Italy”
},
{
“name”: “Emily”,
“activity”: “shared travel stories about a recent trip to Italy”
},
{
“name”: “Anna”,
“activity”: “”
},
{
“name”: “Tom”,
“activity”: “”
},
{
“name”: “Rachel”,
“activity”: “brought a board game that kept everyone entertained”
}
]
}

That is a simple example where the names are associated with activities.

But what if the data/text is not as simple as two columns and several rows?

What if we need to extract complex relationships, quantities, and costs?

Consider this text:

“Yesterday, Sarah went to the local market to buy groceries. She purchased 3 apples, 2 cartons of milk, and a loaf of bread. The total bill came to $12.50. After shopping, she met her friend Mike at a nearby café and had coffee together.”

An LLM can infer structure and produce something like this:

{
“person”: “Sarah”,
“activities”: [
{
“action”: “went to the market”,
“purpose”: “buy groceries”,
“items_purchased”: [
{ “item”: “apples”, “quantity”: 3 },
{ “item”: “milk”, “quantity”: 2, “unit”: “cartons” },
{ “item”: “bread”, “quantity”: 1, “unit”: “loaf” }
],
“total_cost”: 12.50
},
{
“action”: “met a friend”,
“friend”: “Mike”,
“location”: “café”,
“activity”: “had coffee together”
}
]
}

Once we have structured data, we can programmatically interact with it in ways that were previously impossible.

For example:

1) A shopping app could calculate the total cost of Sarah’s groceries and compare it with historical prices.
2) A chatbot could schedule a reminder for Sarah’s next grocery trip.
3) A business analytics tool could track consumer behaviour based on shopping habits.
4) A financial assistant could categorize spending and suggest a budget.
5) Etc.

What this then allows you to do is do other actions that require structured data to take an action, like???

That is one of the most magical parts of Large Language Models that makes it truly transformative, not some magic human that will do everything for you without thinking.

It’s one piece of that puzzle.

###

If you liked this and want to support me but don’t know how, you can buy me a virtual coffee here to say thanks: https://www.buymeacoffee.com/orren or https://ko-fi.com/orren