The definitive glossary of AI terms for indie hackers

Each new week seems to bring a new AI-related term that we all have to learn: “Chain of thought.” “Vibe coding.” “Model Context Protocol.” And so on.

Like it or not, being an indie hacker in 2025 means you have to keep up with all this AI jargon — whether you're a coder or non-technical.

So here's a glossary of all the AI terms you're likely to come across in the indie hackers community on any given week. (Phrases in bold have their own full definition in the glossary.)

Artificial General Intelligence (AGI)

Hypothetical AI systems capable of understanding, learning, and performing any intellectual task that a human can. Unlike current specialized AI, AGI would demonstrate versatile intelligence across diverse domains.

Artificial Intelligence (AI)

Computer systems designed to perform tasks typically requiring human intelligence, such as recognizing patterns, making predictions, or generating content. Unlike AGI, current AI is specialized, performing well only within specific domains.

AI Agents

Autonomous systems or programs capable of performing tasks, making decisions, or taking actions independently based on inputs, objectives, and environmental context. They often combine multiple AI techniques, such as machine learning and reasoning, to achieve specific goals.

AI Alignment

Ensuring AI systems' goals are aligned with human values and intentions. This includes both technical approaches to make AI systems behave as intended and philosophical considerations about what values we should instill in AI systems.

AI Model

A specific algorithm or system trained to perform a particular task within the broader field of artificial intelligence. While AI refers to the overall concept of machines mimicking human intelligence, an AI model is the practical implementation — like a chatbot that generates text, an image generator, or a recommendation engine — built using data and trained to operate within a defined scope.

AI Wrapper

Think of an AI wrapper like a friendly middleman between you and a complex AI model. Instead of dealing with all the complicated code and technical details, the wrapper makes it easy to send questions or commands to the AI and get back clean, understandable answers — kind of like how a phone app makes using the internet easier without needing to know how Wi-Fi works.

Chain of Thought (CoT)

A step-by-step method used by reasoning language models where the AI explicitly outlines each logical step or intermediate reasoning stage leading to a final conclusion or answer. In reasoning models, using CoT enhances transparency, improves accuracy, and allows humans to understand how the model arrives at its decisions.

Chatbot

An AI-powered program designed to simulate human-like conversations, typically using natural language processing (NLP) to understand and respond to text or voice inputs. Modern AI chatbots, like those based on large language models (LLMs), can generate context-aware, dynamic responses, making them useful for customer support, virtual assistants, and interactive applications.

Compute

The processing power required to run AI models, handle data, and perform complex calculations. In AI, compute refers to the hardware resources — like graphics processing units (GPUs) and tensor processing units (TPUs) — that enable models to be trained and run efficiently. More compute generally means faster and more powerful AI capabilities.

Computer Vision

AI that interprets visual data like images or videos. Applications include facial recognition, medical image analysis, autonomous vehicles, and quality control in manufacturing.

Context

The information an AI model retains and considers when generating a new response or making a decision. In large language models, context includes previous words, sentences, or even entire conversations, helping the AI understand meaning, maintain coherence, and produce relevant outputs. The amount of context a model can handle is often limited by its “context window,” which determines how much prior information it can “remember” at any given time.

Deep Learning

A subset of machine learning using neural networks with many layers to identify complex patterns. Deep learning has revolutionized the AI field by enabling breakthrough performance in tasks like image recognition, language translation, and game playing.

Embedding

A numeric representation of text that captures semantic meaning, used by AI models for understanding context. Embeddings allow AI systems to understand relationships between words and concepts, enabling more sophisticated language understanding and generation.

Explainability

The degree to which AI decisions can be understood by humans. This is crucial for high-stakes applications like healthcare and criminal justice, where understanding how AI reaches its conclusions is essential for trust and accountability.

Fine-tuning

Refining a pre-trained model on specific data to improve performance on specialized tasks. This process allows organizations to leverage powerful general models while adapting them to specific domains or use cases.

Foundation Model

Large AI models trained on vast datasets that can be adapted for various specialized tasks. These AI models serve as a base for many applications and can be fine-tuned for specific purposes, dramatically reducing the resources needed for developing specialized AI systems.

Generative AI

AI systems that create original content (text, images, audio), such as ChatGPT. These systems learn patterns from training data and can produce new content that maintains similar patterns while being unique and contextually appropriate.

Graphics Processing Unit (GPU)

A specialized processor originally designed for rendering graphics but now widely used in AI and machine learning because it can handle multiple calculations in parallel. GPUs are essential for training and running AI models efficiently, especially deep learning models.

Ground Truth

Imagine you’re studying for a test, and you have an answer key with all the correct answers — that’s your ground truth. In AI, ground truth is the real, verified data that the AI model learns from, like showing a model thousands of labeled pictures of cats so it knows what a cat actually looks like. It’s how we make sure AI isn’t just guessing but learning from facts.

Hallucination

When generative AI produces plausible-sounding but factually incorrect information. This is a significant challenge in deploying AI systems, particularly in contexts where accuracy is crucial, such as healthcare or legal applications.

Inference

Using a trained AI model to make predictions on new data. The inference process involves applying the patterns learned during training to new situations, often requiring significant computational resources for large models.

Large Language Model (LLM)

AI models trained on massive amounts of text data to generate human-like text (e.g., GPT-4). These models can understand context, generate coherent responses, and perform a wide range of language-related tasks, from translation to coding.

Machine Learning (ML)

A subset of AI involving algorithms that improve through experience and data. ML systems can automatically identify patterns in data and use these patterns to make predictions or decisions without being explicitly programmed with rules.

Model Context Protocol (MCP)

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic, that aims to provide a standardized way for Large Language Models (LLMs) to access and interact with external data sources and tools. It facilitates seamless integration by acting as a bridge between LLM applications and their data, allowing AI to work with context from various sources.

Natural Language Processing (NLP)

AI field focused on enabling computers to understand, interpret, and generate human language. NLP technologies power applications like machine translation, chatbots, sentiment analysis, and automated content generation.

Neural Network

A computational model inspired by the human brain, composed of interconnected nodes (neurons). These networks can learn complex patterns by adjusting the strength of connections between neurons based on training data.

Parameters

Internal variables within AI models that are adjusted during training to shape performance. The number of parameters often correlates with a model's capability, with modern large models containing billions or even trillions of parameters.

Prompt Engineering

Crafting effective instructions or inputs to guide generative AI systems. This involves understanding how to structure prompts to achieve desired outputs and handle edge cases effectively.

Reasoning Language Model

An AI model designed to logically analyze information, draw conclusions, and make decisions based on input data and predefined rules or learned patterns. These models differ from purely predictive models by explicitly following logical processes or steps to arrive at an answer or solution.

Reinforcement Learning

Training AI through rewards and penalties based on actions taken. This approach enables AI systems to learn optimal strategies through trial and error, similar to how humans learn from experience.

Retrieval-Augmented Generation (RAG)

A hybrid AI approach that combines retrieval (fetching relevant information from external sources) with generation (producing responses using a language model). In RAG-based systems, the AI first retrieves real-time or factual data from a database, documents, or the web before generating an answer. A good example is Perplexity AI, a chatbot that actively searches the web for relevant sources and cites them in its responses, combining retrieval with AI-generated summaries.

Supervised Learning

Training AI models with labeled data (inputs and correct answers). This is the most common form of machine learning, where the model learns to map inputs to known outputs based on examples.

Tensor Processing Unit (TPU)

A custom-built hardware component developed by Google, designed specifically for machine learning workloads. TPUs are optimized for tensor computations, making them faster and more efficient than GPUs for certain AI tasks, like training large neural networks.

Tokenization

Breaking text into smaller units (words or tokens) for processing by NLP models. The choice of tokenization strategy can significantly impact model performance and efficiency, particularly for handling multiple languages or specialized vocabulary.

Training

Teaching an AI model by adjusting its parameters based on data. This process requires significant computational resources and carefully curated datasets to achieve optimal performance.

Transformer

A neural network architecture using attention mechanisms, essential for large language models. Transformers revolutionized NLP by enabling models to process long sequences of text more effectively by focusing on relevant parts of the input.

Unsupervised Learning

Training AI models on unlabeled data to discover patterns without explicit instructions. This approach is particularly valuable when labeled data is scarce or expensive to obtain, and for discovering novel patterns in data.

Vibe Coding

When a programmer describes a problem in a few sentences as a prompt to a large language model tuned for coding. This approach can significantly speed up development by automating routine coding tasks and providing quick solutions to common programming challenges.

Weights

Numerical values used within neural networks that determine how strongly each input influences the output. These weights are adjusted during training to help the AI accurately interpret data and make predictions.