AI Slang
Vocabulary from the AI era — from "slop" to "agentic".
50 terms
When an AI model confidently generates false information — citing non-existent sources, inventing pe…
Describing AI systems that can plan, take actions, and use tools autonomously across multi-step task…
Cross-Origin Resource Sharing. The browser security feature that blocks your frontend from calling A…
Accumulated cost of shortcuts taken in code. Like financial debt, it compounds — the longer you wait…
Derogatory slang for an AI system, robot, or automated agent — especially when it fails spectacularl…
Low-effort AI-generated content flooding the internet — generic images, soulless blog posts, fake pr…
Building software by describing what you want to an AI and accepting whatever it produces — often wi…
Application Programming Interface. A contract that lets one program talk to another — endpoints, par…
Anthropic's CLI that turns Claude into an agentic pair-programmer with filesystem, shell, and web ac…
The craft of writing inputs to LLMs to reliably produce the output you want. Includes role-setting, …
Large Language Model. The AI systems (GPT, Claude, Gemini) trained on massive text corpora that powe…
Attack vector where untrusted input (a webpage, a PDF, a tool output) contains instructions that tri…
Excessive, sycophantic praise from an AI assistant — "What a brilliant question!" style. OpenAI famo…
A call to release code to production. Prioritizes getting in front of users over chasing perfection.…
A dismissive reply implying someone's problem is their own lack of skill, not the tool or system. Cl…
Retrieval-Augmented Generation. A pattern where an LLM pulls relevant documents from a vector databa…
"Looks Good To Me." Shorthand approval on a pull request or code review, signaling the reviewer has …
Minimum Viable Product. The simplest version of a product that validates a core hypothesis with real…
Debugging by explaining a problem aloud to an inanimate object — traditionally a rubber duck, nowada…
A hypothetical AI that surpasses the cognitive performance of the smartest humans across virtually e…
The maximum amount of text an LLM can "see" at once, measured in tokens. Longer windows let models p…
A representation an AI builds of how the world works — its physics, objects, causality — that lets i…
Intentional misspelling of "get good." A dismissive response to someone complaining about difficulty…
Wasting disproportionate time debating trivial details (button colors, variable names) while ignorin…
Synthetic media — usually video or audio — where AI convincingly impersonates a real person. Went fr…
Running a trained AI model to produce output. Training teaches the model; inference is every time yo…
The ability of an LLM to call external functions (search, code execution, databases) mid-conversatio…
Continuing to train a foundation model on a smaller, specialized dataset so it specializes in your t…
Performing a seemingly pointless preliminary task that leads to other preliminary tasks — all in the…
An LLM trained or configured to produce long chains of internal "thinking" tokens before answering —…
Mixture of Experts. A model architecture where only a fraction of parameters activate per token — le…
When an LLM's context window fills up with too much scattered history and it starts losing track of …
Agent-to-Agent. Protocols for AI agents to communicate with each other autonomously — emerging as th…
Short for "evaluation." A benchmark or test suite used to measure how well an AI model performs. "Wr…
A database optimized for storing and searching embeddings — the numerical representations that power…
Giving an LLM a handful of input/output examples in the prompt before asking it to handle a new case…
Replacing a running component (a model, a service, a server) without downtime. In the AI era: swappi…
What happens when AI models are trained on the output of earlier AI models — quality degrades genera…
The highest-priority instruction given to an LLM at the start of a conversation. Defines its role, t…
When an AI fabricates plausible-sounding but false information in a confident, high-quality register…
A prompt that tricks an LLM into bypassing its safety training — making it output things it was spec…
The max number of requests you can send an API in a given window. Hitting it returns a 429; every de…
The unit LLMs read and produce — roughly 3/4 of a word in English. Context windows, pricing, and rat…
Model Context Protocol — Anthropic's open standard for connecting LLMs to tools and data sources. St…
"AI slop" — low-effort AI-generated content flooding the internet: generic images, soulless blog pos…
Training a smaller, cheaper model to mimic the output of a larger one. "The 1B distillation of our 7…
An LLM setting that controls randomness. Temperature 0 = deterministic, always pick the most likely …
Shrinking a neural network's weights to lower precision (float16 → int8 → int4) to run it on cheaper…
A massive, general-purpose AI model trained on broad data, from which smaller specialized models are…
"State Of The Art." Used in AI paper titles and benchmark claims — often with self-seriousness. "New…