A-D

AGI (Artificial General Intelligence): Hypothetical AI that can perform any intellectual task a human can. Does not exist yet. Current AI is “narrow” – excellent at specific tasks but unable to generalize across domains.

AI Agent: An AI system that can take actions autonomously, such as browsing the web, writing code, or managing files, to accomplish a goal. Goes beyond simple chat by executing multi-step workflows independently.

Alignment: The challenge of ensuring AI systems behave according to human intentions and values. A major focus of AI safety research.

API (Application Programming Interface): A way for software to communicate with AI models programmatically. Instead of using a chat interface, developers send requests to an API and receive AI responses in their own applications.

Attention Mechanism: The core innovation behind modern AI. Allows models to focus on the most relevant parts of input data when generating output. When translating a sentence, attention helps the model connect each output word to the most relevant input words.

Bias: Systematic errors in AI output caused by unrepresentative training data or flawed model design. Can lead to unfair outcomes for certain groups.

Chain-of-Thought (CoT): A prompting technique where you ask AI to show its reasoning step-by-step. Significantly improves accuracy on complex problems like math, logic, and analysis.

Classification: An AI task where the model assigns inputs to predefined categories. Examples: spam vs. not spam, positive vs. negative sentiment, support ticket priority levels.

Context Window: The maximum amount of text an AI model can process at once, measured in tokens. GPT-4 Turbo handles 128K tokens (roughly 96,000 words). Larger context windows allow processing longer documents.

Diffusion Model: The AI architecture behind image generators like DALL-E, Midjourney, and Stable Diffusion. Works by learning to remove noise from images, then generating new images by starting with pure noise and progressively refining it.

E-H

Embeddings: Numerical representations of text, images, or other data that capture meaning. Similar items have similar embeddings. Used for semantic search, recommendations, and clustering. Think of it as converting words into coordinates on a map where similar concepts are nearby.

Few-Shot Learning: Teaching an AI model to perform a task by providing just a few examples in the prompt, rather than extensive training. Example: showing 3 product descriptions and asking the AI to write more in the same style.

Fine-Tuning: Retraining a pre-existing AI model on your specific data to improve its performance for your use case. Like taking a general-purpose chef and training them specifically in your restaurant’s cuisine and style.

Foundation Model: A large AI model trained on broad data that serves as a base for many applications. GPT-4, Claude, and Gemini are foundation models. They can be used directly or fine-tuned for specific tasks.

Generative AI: AI that creates new content (text, images, code, music, video) rather than just analyzing existing data. ChatGPT, DALL-E, and Midjourney are generative AI tools.

GPU (Graphics Processing Unit): The hardware that powers AI model training and inference. Originally designed for gaming graphics, GPUs excel at the parallel math operations AI requires. NVIDIA dominates this market.

Guardrails: Rules and constraints placed on AI systems to prevent harmful, inappropriate, or off-topic outputs. Can be implemented through system prompts, content filters, or output validation.

Hallucination: When an AI model generates information that sounds plausible but is factually incorrect. A fundamental limitation of current generative AI. Always verify critical facts from AI output.

I-P

Inference: The process of running an AI model to generate output. When you send a prompt to ChatGPT and receive a response, that is inference. Distinct from training (building the model).

LLM (Large Language Model): An AI model trained on massive text datasets that can understand and generate human language. GPT-4, Claude, Gemini, and Llama are LLMs. “Large” refers to billions of parameters (adjustable weights in the model).

LoRA (Low-Rank Adaptation): An efficient fine-tuning technique that modifies only a small fraction of a model’s parameters, making customization faster and cheaper. Enables fine-tuning on consumer hardware rather than data center GPUs.

Machine Learning (ML): A subset of AI where systems learn patterns from data without being explicitly programmed. Traditional ML uses structured data and algorithms like random forests and gradient boosting. Deep learning (neural networks) is a subset of ML.

Multimodal: AI that processes multiple types of input (text, images, audio, video) simultaneously. GPT-4V, Gemini, and Claude 3 are multimodal models that can analyze images alongside text.

NLP (Natural Language Processing): The branch of AI focused on enabling computers to understand, interpret, and generate human language. Encompasses tasks like translation, summarization, sentiment analysis, and question answering.

Open Source (AI): AI models whose weights and architecture are publicly available for anyone to use, modify, and deploy. Llama (Meta), Mistral, and Stable Diffusion are prominent open-source AI models.

Parameters: The adjustable weights in a neural network that determine the model’s behavior. GPT-4 has an estimated 1.7 trillion parameters. More parameters generally (but not always) means more capability.

Prompt Engineering: The skill of crafting effective instructions for AI models to produce desired outputs. Involves specifying context, constraints, format, and tone. A critical skill for getting value from generative AI.

R-Z

RAG (Retrieval-Augmented Generation): A technique that combines AI generation with real-time information retrieval. Instead of relying solely on training data, RAG systems search a knowledge base for relevant information and include it in the AI’s context. Reduces hallucinations and keeps responses current.

Reinforcement Learning from Human Feedback (RLHF): A training technique where humans rate AI outputs and those ratings are used to improve the model. This is how ChatGPT was trained to be helpful and safe. Human feedback teaches the model what “good” responses look like.

Semantic Search: Search that understands meaning rather than just matching keywords. Searching for “affordable places to stay in Paris” would match results about “budget hotels in Paris” even without exact word overlap.

SLM (Small Language Model): Compact AI models (under 10 billion parameters) optimized for specific tasks or edge deployment. Run on laptops and phones. Examples: Phi-3 (Microsoft), Gemma (Google). Trade breadth for efficiency.

System Prompt: Instructions given to an AI model that define its persona, capabilities, and constraints before user interaction begins. Sets the rules of engagement for the entire conversation.

Temperature: A setting that controls AI output randomness. Low temperature (0.1-0.3) produces predictable, focused responses. High temperature (0.7-1.0) produces creative, varied responses. Use low for factual tasks, high for brainstorming.

Tokens: The units AI models use to process text. Roughly 1 token = 0.75 words in English. A 1,000-word article is approximately 1,333 tokens. Pricing and context limits are measured in tokens.

Transformer: The neural network architecture behind modern AI language models. Introduced in Google’s 2017 “Attention Is All You Need” paper. Uses self-attention mechanisms to process sequences of data in parallel rather than sequentially.

Vector Database: A specialized database that stores and retrieves embeddings (vector representations of data). Essential infrastructure for semantic search, recommendations, and RAG systems. Examples: Pinecone, Weaviate, Chroma.

Zero-Shot Learning: When an AI model performs a task it was not specifically trained for, using only the task description in the prompt. Example: asking a general AI model to classify customer feedback by department without providing examples.

Related Articles

Frequently Asked Questions