🔤 Acronym Decoder

The US standard for registered business texting over normal 10-digit phone numbers.

A second proof of identity beyond your password — like a code from an app.

An open protocol that lets separate AI agents discover each other and collaborate on tasks.

Business messaging sent from software to people. In the US it requires 10DLC registration.

A hypothetical AI that can do any intellectual task a human can, across all domains. It doesn't exist yet.

Computer systems that perform tasks normally needing human intelligence — understanding language, recognizing images, and making decisions.

The formal name for a neural network — software 'neurons' in layers that learn from examples.

A defined way for two pieces of software to talk and exchange data — the backbone of every integration.

A theoretical AI that surpasses the smartest humans at virtually everything. Still hypothetical.

The technology behind speech-to-text — machines transcribing what people say.

The rules for what you're allowed to do with a product or API.

Selling products or services to other companies.

Selling directly to individual people.

A network of servers that caches your site near users so it loads fast worldwide.

Automated pipelines that test and ship your code on every change.

Controlling software by typing commands instead of clicking — how tools like Claude Code and git work.

A neural network great at images — it detects edges, shapes, and patterns. Common in computer vision.

Prompting a model to reason step by step before answering, which improves accuracy on hard problems.

Software to manage leads and customers — GoHighLevel is one.

The four basic operations on stored data — the core of most apps.

A plain-text table format — rows of values separated by commas. Opens in any spreadsheet.

AI that interprets images and video — detecting objects, faces, or text.

An organized store of data your app reads from and writes to.

A digital signature proving an email really came from your domain and wasn't tampered with.

A branch of ML using many-layered neural networks to learn complex patterns. It powers modern AI like LLMs and image generation.

A policy that tells receivers what to do when an email fails SPF or DKIM checks.

The internet's phone book — it turns domain names into the server addresses behind them.

A simpler, popular way to align a model toward preferred answers without full reinforcement learning.

The process of pulling data, cleaning it, and loading it into a destination.

A measure of how much raw computation a model uses. More FLOPs means more compute — and cost.

Two networks competing — one generates fakes, one detects them — used to create realistic images.

AI that creates new content — text, images, audio, video, or code — rather than just classifying or predicting.

An all-in-one CRM and marketing platform built for agencies — your main stack.

The model architecture behind many LLMs — 'pre-trained' on huge text, then used to generate more.

A chip that does many calculations in parallel — the workhorse for training and running AI.

Your plan for launching and selling a product.

Anthropic's shorthand for how a well-aligned assistant should behave.

A design where a person reviews or approves an AI's actions before they take effect.

The protocol browsers and apps use to talk to web servers.

HTTP with encryption, so data in transit stays private.

Animating a still image into a short video.

Renting raw cloud computing — servers and storage — like AWS EC2.

A model learning a task purely from examples in the prompt, with no retraining.

An app for writing code with built-in editor, debugger, and tools — like VS Code or Cursor.

A simple text format for structured data — how most APIs and webhooks exchange information.

A measurable number that tracks progress toward a goal.

A speed trick that stores past attention data so a model doesn't recompute it for every new token.

An AI trained on massive amounts of text so it can understand and generate human language. It's the tech behind ChatGPT, Claude, and Gemini.

A model that handles more than text — images, audio, or video — in one system, like describing a photo you upload.

A popular PEFT method that adds tiny trainable layers instead of retraining the whole model.

Anthropic's open standard that lets AI apps connect to tools and data — like a USB-C port for AI.

Two or more verification steps for stronger account security.

A type of AI where systems learn patterns from data instead of being hand-programmed with rules.

A popular benchmark testing a model's knowledge across 57 subjects.

Text messages that include images, audio, or video.

A model design that routes each input to a few specialized 'expert' sub-networks, giving big-model quality at lower cost.

The simplest version of a product with only the essential features needed to solve a real problem and gather feedback.

The part of NLP focused on producing human-like text.

The field of getting computers to understand and work with human language.

The part of NLP focused on grasping the meaning and intent behind text.

A computing system loosely inspired by the brain, made of connected 'neurons' that learn from data.

A standard that lets apps access your account without you sharing your password.

Turning images of text (scans, photos, receipts) into editable, searchable text.

A tool that lets you use a database from code without writing raw SQL (e.g., Prisma).

Software whose code is public and free to use, modify, and share.

A short code sent to verify identity, usually for login or checkout.

Cloud platforms that run your app for you, without managing servers (e.g., Vercel, Railway).

The craft of writing instructions that reliably get the best output from a model.

Fine-tuning only a small slice of a model's weights to save time and cost.

Data that identifies a person (name, email, phone). Handle it carefully and never log it carelessly.

A small build that proves an idea is technically possible before investing more.

LoRA applied to a compressed (quantized) model, so you can fine-tune large models on modest hardware.

Work focused on creating new products or improving existing ones.

Giving a model relevant documents at query time so it answers from your data, not just its memory.

Practices for building AI that's fair, safe, transparent, and accountable.

A pattern where a model alternates thinking and taking actions (like calling tools) to solve a task.

A common style for web APIs using standard HTTP methods like GET and POST.

Training where an agent learns by trial and error, earning rewards for good actions.

Like RLHF but using AI-generated feedback instead of humans, to scale alignment cheaply.

Fine-tuning a model using human ratings of its answers, so it behaves more helpfully and safely.

A neural network for sequences like text or time series that remembers previous steps. Mostly replaced by transformers.

How much value you get back compared to what you put in.

Software you use over the internet by subscription — like GoHighLevel.

A bundle of tools and libraries that makes building on a platform (like the Claude API) much easier.

Training a base model on labeled examples to specialize it for a task, tone, or format.

A promise about service quality — like guaranteed uptime or response time.

A compact language model that runs cheaper and faster — even on a phone or laptop — trading some capability for speed and privacy.

Standard text messaging between phones.

A documented step-by-step process for a repeatable task.

A DNS record listing who's allowed to send email for your domain — helps you avoid spam folders.

The language for querying and managing databases.

Logging into many apps with one set of credentials.

Converting spoken audio into written text — used for transcription and voice commands.

A benchmark that tests how well AI fixes real software bugs — a key scoreboard for coding models.

Generating images from a text prompt, like Midjourney or DALL·E.

Generating video clips from a text prompt.

The legal agreement that governs how you may use a service.

An advanced method where the model explores multiple reasoning paths and picks the best one.

Google's custom AI chip, optimized for the math machine learning relies on.

Converting written text into spoken audio — the tech behind AI voices.

The visual part of an app that users see and interact with.

A web address that points to a page or resource.

How it feels to use a product — the flow, clarity, and ease.

The memory on a GPU. It limits how big a model you can run or fine-tune.

Techniques that make an AI's decisions understandable to humans instead of a black box.

A human-friendly format used for configuration files.