The Lakes Every Reporter Should Know: Data Lakes

Large Language Models (LLMs) come from data lakes. These are big pools of unstructured data. Think about video, tweets, tik tok, what's app, emails, podcasts, or radio. Meta data is continuously flying around. All this vast information gets stored in data lakes. LLMs like Chat GPT use that collection of assorted information, and based on a set of rules, can generate words and sentences it has been taught to produce.

When we start using LLMs what we are doing is interacting with automated intelligence, something that knows the rules of sentence structure and grammar. If this, then that. The answers that get generated are simply a reflection of the data set it has been fed, and what was available at the time, at the geographical location a person made their query. It's not necessarily factual.

Candyce Kelshall, President of the Canadian Association of Security and Intelligence Studies, tells me "this is scary, it's as basic as sweeping the floor" ,that's how information is just scooped up. Journalists need to think of who is behind the data they're replying on, while keeping pace with technologies, she says.

"From a security perspective.When you go home you lock your door, you close your curtains. You use a trusted provider to order food. You're careful about the internet. The same rules apply when it comes to information management in a newsroom."

To ensure data hygiene, the kind of people news leaders hire into a newsroom is important. The right skill sets and talent need to be in place. It's important to know who AI software developers are. What kind of products they create. Look for the track record behind all the AI that's used because all code is vulnerable.