A simpler explanation of how AI chatbots work

AI-powered chatbots are among the fastest adopted technologies in human history. They jumpstarted a new focus on artificial intelligence.

It helps to better understand how they work. Right now, the dominant method of powering chatbots that generate human-like language is large language models, or LLMs. OpenAI’s ChatGPT and others like Claude, Google’s Gemini and DeepSeek all use LLMs, which are computer systems that have been trained on enormous collections of human words — from books to social media postings to video transcripts.

All this is important but complex, especially for those outside of computer science. We at Technical.ly speak most often to our audience of entrepreneurs and technologists who are deeply involved here, but as a news org, we feel especially able to help translate between those most experienced with technology, and those eager to catch up.

So, building on our existing Creator in Residence program, we’re intending to work with other individual creators to interpret our reporting and other technical terms for wider audiences. 

First up, I got together with Pop Pop Bruce Johnson, the folksy influencer with a million followers on TikTok and more than half a million on Instagram, to better describe LLMs.

As Pop Pop and I put it: 

Think of them like super-smart talking parrots. They sound like they’re talking like humans, but they’re just mimicking sounds. 

LLMs are remarkable probability machines, predicting the most likely next word or phrase to mimic human writing. Revolutionized by an important 2017 paper, LLMs are extraordinary, but by no means perfect. 

Their “hallucinations”— which AI companies are working to reduce — are a consequence of how they work: They don’t “understand” what a person is asking, they just predict the most likely response.

For what it’s worth, LLMs are only one kind of way to power an AI tool. For example, most image generation tools use something called a diffusion model, instead of a large language model (LLM). The rise and fall of LLMs may, or may not, mirror AI more generally.

Better understanding how these and other bits of technology work help make us better able to navigate their impact.