How does AI work and why do I care?
The benefits of better understanding how AI systems function and how chatbots answer your questions.
FROM: Ben Lovell
RE: How does AI work and why do I care?
QUESTION PRESENTED:
Does it benefit me to better understand how AI chatbots answer my questions?
SHORT ANSWER:
AI Models are predictive machines. Understanding how these models analyze input and generate answers will help you scrutinize their output. Ask the same question in different ways and get significantly different answers. How you ask your questions matters.
DISCUSSION
I. Can I trust AI models?
No. No you cannot. As stated above, Large Language Models run entirely on vector multiplication and multidimensional manifolds predicting the next word based on the context of all the other words you input into the model. The model completes the story.
This works wonderfully for confirmable tasks, like math problems with a known solution, but not so great for generating persuasive arguments in a legal context. As a lawyer, clients trust me for my judgment and experience, therefore I cannot delegate that trust to an AI model.
II. Models, how do they work?
Large Language Models (LLMs) predict text by multiplying and processing incredibly large collections of numbers representing connections in higher dimensional manifolds.
The LLM breaks down your question into tokens (collections of numbers representing the words you used) and performs functions on those matrices. Once the math is done the LLM outputs more tokens representing the most likely completion of the provided input. Wash, rinse, repeat in parallel. The inner workings and math of it all are, to me, extremely complicated.
Visualization of the embedding graph for the sentence “AI agents will be the most hot topic of artificial intelligence in 2025,” from “LLM Embeddings Explained: A Visual and Intuitive Guide”, 2025.1
III. So, what can I use AI for?
LLMs are great for searching PDFs or large data sets and finding connections. They are skilled at outlining legal requirements for a motion. They make the discovery process easier.
Two examples:
- Create deadlines from a scheduling order.
- Extracting deadlines from a document is a great use case for an LLM. In your prompt, you need to tell the LLM what a scheduling order is, how deadlines are presented in the document, how to calculate any relative deadlines, and the rules for how weekends are handled in your jurisdiction.
- Outline motions
- Legal arguments are outlined formulaically from statute or case law and LLMs can provide a good start. Ask an LLM to outline a motion by section based on your statutory requirements and work from there.
IV. What can’t I use AI for?
I do not use AI to write. I do not use LLMs for anything beyond basic outlines. Not only will offloading your critical thinking to AI make you worse at your job, it will increase your propensity to make a mistake.