Quick AI Glossary
The other day, after publishing a post about an AI competition I participated in, my brother told me that he didn't fully understand several terms. He said that if he couldn't grasp them, probably my audience couldn't either. So I decided to make this small guide with concepts that I sometimes take for granted that everyone knows.
LLM (Large Language Model)
It's an algorithm capable of understanding and writing in human language. It's important not to confuse it with ChatGPT; ChatGPT is an application that uses an LLM (currently, a model like GPT-5). LLMs are trained by reading enormous amounts of internet content, books, and other text sources to learn to communicate like us. They work using probabilities to predict what is the best word that follows in a sentence. Today there are many on the market (GPT-4o, Claude 3, Llama 3, Grok, etc.), each with their strengths and weaknesses.
Prompt (Instruction)
The prompt is simply the text or instruction you send to the LLM to ask it to do something. It's your starting point for the conversation.
Prompt example
Prompt Engineering
These are the techniques used to write better prompts. It's more than proven that a well-structured prompt, with good context and clear details, produces much better and more precise results from LLMs. (Since I'm writing more often, I have in my "backlog" a guide of the techniques I use to improve my prompts)
Token
Tokens are the way LLMs process language. Instead of "seeing" complete words, they divide the text into smaller pieces (tokens) to understand it better. For example, "walking" could be "walk" + "ing". Every time you use an LLM, the cost is calculated based on input tokens (your prompt) and output tokens (its response).
tokens
Fun fact: Since it's predicted that AI could automate many jobs, there are already important people, like Anthropic's CEO, who have proposed the idea of charging a "token tax" in the future.
AGI (Artificial General Intelligence)
AGI (Artificial General Intelligence) is the great dream of many in this field (myself included). It's a theoretical concept that describes an AI that would be better than any human at practically all tasks. Imagine an AI that's a better engineer than me, better lawyer than Eugenio, better doctor than Dr. Pérez, better president than President Boric, better soccer player than Messi and so on in every job.
AI Agent (Artificial Intelligence Agent)
An agent is a "program with its own initiative" that not only responds to what you ask it, but can take actions on its own based on what it detects and what it knows, until it fulfills its objective. Right now the term agent is in a hype phase, where many call an AI script that answers questions an agent (I've done it too), when in reality the promise is much bigger
In theory, in the coming years we're going to see agents that should be capable of doing things like:
- Collaborate digitally with you (joins a meeting, writes a document, makes a purchase, etc)
- Make decisions on its own, even when you're not watching.
- Coordinate resources and tools to fulfill its mission
- Learn and improve over time
Right now most (if not all) "agents" don't have:
- Long-term memory
- Ability to coordinate complex tasks
- Robust reasoning
- Learning capability
How I imagine an agent in the future
Bias
It's a prejudice the model has. Because LLMs learn from texts written by humans, they inherit our biases: they can be racist, sexist, or xenophobic (just like us, sometimes). In some cases, these biases are included intentionally, as in the case of Grok, which often reflects Elon Musk's positions.
Hallucinations
This is when AI makes up information and presents it as if it were a real fact. Be very careful with this! LLMs sometimes generate data, quotes, or events that never happened. It's always a good idea to verify important information from other sources.
Benchmark
It's a performance test given to LLMs to see how they perform against different challenges. There are benchmarks to measure text comprehension, mathematical reasoning, general knowledge, and many other skills.
These are the terms I have for now. I'm going to keep updating this list of terms as I write more entries on my blog.