AI Empathy

Education leads to empowerment. Consider this viral technology through the lens of its scientific definition — large language model — and let it inform the way you use the technology.

Let’s get our bearings — imagine you are floating in space, surrounded by stars. Now imagine all of those distant white dots being brought into your bedroom. Each white dot represents an idea or a thing. Ideas like cat, dog, hamburger. Each white dot has an arrow that represents a relationship — cats and dogs have arrows that point at “fluffy fur,” hamburgers point at “fast food.” Cats and dogs also point to “tail” and “ears” and those are shared by other animals. The majority of arrows pulls your eyes in a direction and you start to see most arrows pointing to a star called “mammal,” which is shared by other animals including “human,” which is being pointed at by star that has your name on it. This is a large language model.

Every concept that has been put into words is a direction in your room. Dogness is the top left corner, catness is near it. Santa Claus is somewhere in the room, too. Pine tree, Europe, and Mary Oliver — there are billions of directions that you can traverse. You can make a poem more dog, less European, and slightly Santa depending on where you are in the room.

This room is a neural network, and each star is a node with coordinates that are measured by its relationships, we call these complex locations “vectors.” When you ask an LLM a question, it looks at each word, each node, and traverses the network toward your answer. It is also moving toward “correct” and “kind” and “personable,” and eventually arrives at a response.

The chain-of-thought “thinking” process is part of the response itself. Not unlike me writing these words right now, I do not know the words I’ll be typing two lines down, but I do know the next word, which will inevitably lead to future words, of which are two lines down, completing this illustration. So instead of producing the definitive answer like a Google Search, your answer is grown as it moves toward vectors in the network.

Perhaps this is the beginning of understanding AI not as magic, nor as a slot machine, but as scientific and mathematical innovation that can be applied instead of abused.