Glossary AI & Vibecoding
Fundamental reference guide for the new era of software engineering by Andriy Orlov: autonomous agents, loop engineering, Model Context Protocol, context scaling, VPS infrastructure, and burnout prevention.
A
ENAgent Skills & Custom Workflows
An architectural pattern for dynamically loading specialized procedural instructions, scripts, and templates (SKILL.md) into an agent's context window on demand (On-Demand Loading).
AI Hallucinations & Confabulations
The generation of factually incorrect, fabricated, or non-existent information (libraries, API methods, quotes) by a language model, expressed with high probabilistic confidence.
AI Translation and Cultural Localization: Why LLMs Outperform Dictionaries
A methodology for artistic translation and cultural adaptation of texts (Localization). Explains why large language models translate idioms, humor, slang, and domain-specific nuances significantly more accurately and naturally than traditional translators.
AST Chunking for Codebases
A methodology for intelligent chunking of code files for vector search exclusively at the syntactic boundaries of programming languages (Tree-sitter) instead of slicing by a fixed number of lines or characters.
C
ENChain of Thought (CoT)
A methodology that prompts a language model to generate sequential intermediate reasoning steps before producing a final answer, converting additional tokens (Test-Time Compute) into quality and accuracy of the output.
Chunking Documents: A Beginner's Guide
The technique of breaking large documents (PDFs, books, long reads) into smaller logical text blocks (chunks of 300–500 tokens) with overlap. This ensures high search accuracy and prevents context loss at the seams.
Claude Projects (Workspaces with Custom Knowledge Base)
A workspace organization tool in Claude by Anthropic. It allows the creation of dedicated workspaces for each task, uploading up to 200,000 tokens of internal documentation, and sharing them with colleagues.
ColBERT & Late Interaction Retrieval
The neural search architecture compares contextual embeddings of each individual query token with document tokens (Late Interaction), surpassing classical Dense vectors in accuracy.
Constrained Decoding & Structured Outputs
A hardware and algorithmic constraint on token generation by language models at the inference level, mathematically ensuring 100% compliance with JSON Schema or Zod types.
Context Distillation & Token Squeezing
Methods for automatically removing excess words, syntactic noise, formatting spaces, and outdated messages before sending prompts to the model, optimizing budget and accelerating inference.
Context Poisoning in Knowledge Bases
A cyber threat to RAG systems where an attacker injects falsified documents, misleading instructions, or hidden prompt injections into a shared knowledge base (Notion, Confluence, Google Drive). During queries, the system retrieves poisoned context, leading the bot to provide harmful advice or disclose secrets.
Context Rot & Attention Decay
Systemic degradation of accuracy, instruction adherence, and logical consistency in LLMs as dialog noise, outdated code drafts, and compiler outputs accumulate in the working context window.
Context Window
The maximum operational token capacity that a language model can simultaneously hold in the Self-Attention mechanism and KV Cache memory during a single inference request.
Context Window Overflow & Compaction
Engineering patterns to prevent catastrophic failures when context limits are reached: Rolling Summaries, state folding, and selective eviction of stale history.
Context Window Size (Current Conversation Memory)
The maximum amount of text (in tokens) that a language model can simultaneously retain in memory during an ongoing conversation. It determines the length of documents that can be loaded at once without loss of content.
Copyright and AI
The legal status of intellectual property for content generated by neural networks. Precedents from the US Copyright Office and EU: lack of copyright on pure generative outputs, the status of training on third-party works (Fair Use), and licenses for open models.
Cosine Similarity
A mathematical metric that measures the angle between two vectors in a multidimensional space. It indicates the degree of semantic similarity between two texts, ranging from -1 (opposite) to +1 (complete synonyms), while ignoring sentence length.
Cross-Encoder Reranking
A two-stage retrieval methodology in RAG systems: a fast initial candidate selection (Bi-Encoder / BM25) followed by precise ranking through a fully-connected cross-encoder model (Cross-Encoder / Cohere Rerank / BGE-Reranker).
D
ENDelimiters and XML Tags (Structuring Prompts)
A technique for formatting complex prompts using XML tags (<context>, <rules>) and triple quotes (\"\"\"). It helps the model clearly distinguish where your rules end and where the text for processing begins.
Dense Retrieval vs Keyword Search
A comparison of two search approaches: neural network-based semantic retrieval (Dense Retrieval) and traditional keyword matching (Sparse / BM25). It explains why vector search may fail in product item searches and how hybrid search operates.
Document Chunking Strategies
A methodology for decomposing massive documents and codebases into information-rich, self-contained fragments (chunks) for generating vector embeddings and precise retrieval in RAG systems.
E
ENEmbeddings Simplified (How Text Becomes Numbers)
A fundamental technology that transforms words, sentences, or images into multi-dimensional lists of numbers (vectors). It enables computers to mathematically measure semantic proximity between different thoughts and concepts.
Episodic vs Semantic Agent Memory
Architectural separation of AI agent memory into a long-term factual knowledge base (Semantic Memory) and a chronological event log of specific working sessions (Episodic Memory).
Explain Code to Me (AI as a Personal Programming Tutor)
A methodology for utilizing artificial intelligence to learn programming and understand foreign code. It breaks down any convoluted script into relatable analogies, comments, and step-by-step explanations.
F
ENFew-Shot Prompting (In-Context Learning)
A prompt engineering methodology where the model adapts to a specific format, style, or logic during inference (In-Context Learning) by showing 2–5 reference examples.
File Upload Analysis (Attachment Icon in Chat)
The file upload feature in ChatGPT, Claude, and Gemini (attachment or plus icon) allows models to analyze Excel spreadsheets, PDF reports, text documents, and images for instant insights and calculations.
G
ENGraphRAG & Knowledge Graph Retrieval
The next generation of augmented generation search systems (GraphRAG) combines semantic vector search with knowledge graphs to synthesize global insights over large knowledge bases.
Grounding Facts: Linking AI Outputs to Verified Sources
A methodology for anchoring AI-generated assertions to verified primary sources (Source Grounding). This approach eliminates hallucinations and ensures legal and scientific credibility through precise citations and references.
H
ENHallucination Risks in Real-World Tasks
An analysis of the practical consequences and legal risks arising from blind trust in fabricated facts and false sources generated by language models in financial reports, legal claims, and medical advice.
Hierarchical Chunking (Parent-Child)
An advanced document slicing strategy (Parent-Document Retriever) for RAG systems. Text is divided into small child chunks for precise vector search, while the entire parent chunk is passed to the model to maintain broad context and preserve meaning.
Hierarchical Chunking & Parent-Child Retrieval
An architectural pattern for retrieval where vector matching occurs on concise Child Chunks, while the entire broad Parent Document is pulled into the model context.
How to Prevent AI Hallucinations (Prompts Against Hallucinations)
A set of proven engineering techniques and verbal constructs that block the generation of fabricated facts, nonexistent laws, and false citations. Forces artificial intelligence to rely strictly on provided sources.
Hybrid Search (Dense + Sparse Search)
The retrieval architecture in modern RAG systems combines semantic vector search (Dense Embeddings) with classical keyword-based full-text indexing (Sparse/BM25) through rank fusion algorithms (RRF).
Hybrid Search: Dense + Sparse / BM25
An information retrieval architecture that combines vector semantic search (Dense retrieval — understanding meaning and synonyms) with classical keyword search (Sparse retrieval / BM25). It provides an ideal balance between conceptual understanding and precise searches for articles, serial numbers, or names.
I
ENJ
ENK
ENKnowledge Base Connectors (Google Drive, Notion, Confluence)
This technology integrates corporate data sources with language models using pre-built connectors (Connectors / ETL). It automatically synchronizes updated documents from Google Drive, Notion, Confluence, and Slack with a vector database without manual file uploads.
Knowledge Graphs in Search (GraphRAG)
The evolution of RAG architecture from Microsoft Research. It combines classical semantic vector search with a Knowledge Graph, transforming documents into entity nodes (people, companies, locations) and edges of relationships, enabling answers to complex global queries across the data set.
KV-Cache Offloading & Compression
Hardware and algorithmic methods for temporarily offloading Key-Value Cache (KV-Cache) from expensive GPU VRAM to system RAM or fast NVMe SSDs.
M
ENMarkdown AST for Agents (Abstract Syntax Tree)
A hierarchical tree-like representation of Markdown markup (mdast / Unified.js) that enables software systems and AI agents to deterministically analyze, transform, and safely edit technical content without fragile regular expressions.
Meta-Prompting: AI Enhances Your Prompt for You
Meta-Prompting is a technique for crafting higher-order queries. Instead of struggling with complex formulations, you delegate the task to a powerful model that writes, structures, and tests the ideal prompt for your task.
Metadata Filtering in RAG
A technique for combined searching in vector databases (Self-Querying Retriever). It allows for the integration of semantic search by meaning with strict database filters based on metadata: publication date, document author, access level, language, or company department.
Multimodal Embeddings (CLIP)
This technology designs various media types (text, images, audio) into a unified shared multidimensional vector space using architectures like OpenAI CLIP or Google SigLIP. It enables searching for images using textual descriptions or finding similar music based on pictures.
N
ENNeedle in a Haystack & Long-Context Retrieval
The degradation of attention in language models within massive context windows (1M–2M tokens) leads to the model ignoring instructions buried within the text, necessitating engineering methods to overcome this issue.
Negative Constraints & Anti-Pattern Prompts
A methodology for formulating prohibitions and explicitly excluding undesirable engineering patterns (Anti-Goals) in system instructions to prevent boilerplate and excessive code (AI Slop).
NotebookLM and Audio Overview (Podcast from Summaries)
A free research tool from Google based on Gemini models. It strictly adheres to uploaded sources (no hallucinations) and can transform dull reports into engaging conversational audio podcasts.
P
ENParsing Complex PDF Documents
Technologies for extracting structured text from complex PDF files (LlamaParse, Unstructured.io, Marker, Nougat). Explains why simple text copying disrupts reading order in two-column documents and how to preserve formulas, charts, and nested tables for RAG systems.
Prompt Caching & KV Cache Reuse
A technology utilized by modern inference engines and cloud APIs (Anthropic, OpenAI, DeepSeek, vLLM) that stores precomputed attention matrices (KV Cache) of static prefixes, reducing processing costs by 80–90% and decreasing time to first token (TTFT) by 4–8 times.
Prompt Caching Architecture & Economics
A methodology for designing prompts focused on maximizing hardware cache hits for prefixes (Prefix Caching) in Anthropic, OpenAI, and DeepSeek, achieving a 90% cost reduction and 80% speedup.
Prompt Chaining Basics for Beginners
A methodology for breaking down complex tasks into a sequence of focused prompts (Prompt Chaining). The output of the first step feeds into the next, ensuring high quality for each segment.
Prompt Engineering (Context Architecture & Prompt Engineering)
An engineering discipline focused on structuring system directives, XML markup, semantic delimiters, and examples to achieve deterministic, predictable outcomes from probabilistic models.
Prompt Extraction & Inversion Attacks
A methodology for analyzing cyber threats aimed at extracting secret system prompts, proprietary business logic, and hidden instructions through manipulative user queries.
Prompt Injection
A critical vulnerability in LLM-based systems (OWASP Top 10 for LLM #1). It arises from the lack of architectural separation between control instructions (Control Plane) and external data (Data Plane), allowing an attacker to hijack model control.
Prompt Leakage
A vulnerability in artificial intelligence where users cleverly phrase requests to compel the bot to quote its hidden instructions (System Prompt) verbatim, revealing business logic, behavioral rules, and internal secrets of developers.
Q
ENR
ENRAG (Retrieval-Augmented Generation)
An architectural pattern for corporate AI that dynamically enriches the model's context window with relevant verified knowledge from external repositories (vector databases, graphs, full-text indexes) before generating the final response.
RAG vs Fine-Tuning (The Eternal Dilemma of AI Implementation)
A fundamental architectural choice for businesses. RAG (Retrieval-Augmented Generation) versus Fine-Tuning (modifying model weights through additional training). Criteria for choosing between factual relevance and specific behavioral style.
Reranking Search Results (Reranking / Cross-Encoders)
A two-step search pattern in RAG systems. The first step (fast vector or hybrid search) retrieves 20-50 candidates in 10 ms. The second step (Cross-Encoder reranker model, such as Cohere Rerank or BGE-Reranker) meticulously matches the query with each retrieved text, retaining the top 3 most accurate documents.
Role Prompting: How Role Assignment Radically Changes Response Quality
A technique for formulating prompts by assigning the model a specific role, profession, or psychological persona (e.g., 'Act as a Senior DevOps', 'Imagine you are a pediatrician'). This activates specific layers of knowledge and professional vocabulary.
S
ENSelf-RAG & Corrective RAG (CRAG)
An adaptive search methodology where the language model autonomously assesses the relevance of retrieved documents, filters out noise, and dynamically reformulates the search query when information is lacking.
Self-Reflective Search (Self-RAG & Corrective RAG)
The architectural approach of Self-RAG and Corrective RAG (CRAG) enables a language model to critically assess the relevance and completeness of retrieved documents before generating a response, avoiding blind reliance on initial search results. If the findings are insufficient or dubious, the system automatically reformulates the query and searches again or accesses the open web.
Semantic Search: Why Meaning-Based Search Outperforms Traditional Ctrl+F
A technology for information retrieval based on the conceptual content of a query rather than exact letter matches or keywords. It understands synonyms, paraphrasing, spelling errors, and abstract natural language questions.
System Instructions (System Prompt and Custom Instructions)
The primary hidden directive from the developer or user (System Message). It establishes fundamental behavioral frameworks, roles, communication styles, and prohibited topics that the model retains throughout the session.
System Prompt (System Instructions & Metaprompting)
The primary metacontext block of instructions passed at the zero position of the context window, defining the agent's role, safety rules, available tools, and behavioral boundaries.
System Prompt Drift & Degradation
The phenomenon of gradual loss of primary instructions, response style, and security constraints by the language model as the dialogue history expands, along with methods for periodic rule reinjection.
T
ENText Summarization (How to Condense 50 Pages into 5 Key Points)
A methodology for effectively condensing large volumes of information (Summarization). It explains how to transform 50-page reports into concise executive summaries using clear prompt templates following the structure: 'Core Insight + Risks + Next Steps'.
The Magic of the Phrase 'Think Step by Step'
A legendary prompt engineering technique (Zero-Shot Chain-of-Thought). Adding the phrase 'Let's think step by step' compels the neural network to unfold a chain of intermediate calculations, reducing logical errors by 50–70%.
V
ENVector Databases (Vector DBs & ANN Search)
Specialized DBMS and extensions (Qdrant, pgvector, Milvus, Chroma, Turso) optimized for storing millions of high-dimensional vectors and ultra-fast Approximate Nearest Neighbors (ANN) search.
Vector Embeddings (Dense Embeddings)
A mathematical projection of text, code, or multimodal data into a dense, multidimensional numerical vector, where the angle and geometry between coordinates reflect their semantic affinity.
Vector Metadata Filtering & Hybrid Routing
This technology combines semantic vector search with strict deterministic SQL/NoSQL filters on fields (tenant_id, version, role, date) before calculating vector distances.
W
ENWeb Search in Chat (Real-Time Internet Search)
An integrated web search feature within ChatGPT, Claude, and Gemini (globe icon). It enables the model to retrieve fresh data from websites for the current day and provide answers with direct links to sources.
Why the Word 'Not' Breaks Prompts (Positive Instructions)
The psychological and algorithmic paradox of language model behavior ('The Pink Elephant Paradox'). This explains why commands containing 'not' often lead to counterproductive outcomes and how to rephrase prohibitions into positive directives.
Z
ENZero-Shot CoT vs Dynamic Reasoning
The historical and practical evolution of reasoning techniques: from the simple phrase 'Let's think step by step' (Zero-Shot Chain-of-Thought) to native computational budgeting of reasoning in modern models.
Zero-Shot vs. Few-Shot (The Power of Examples)
A comparison of two fundamental prompting techniques. Zero-Shot involves a direct query without examples, while Few-Shot provides the model with several samples of the desired response before executing the task for precise calibration of format and tone.
Full Stack of Tools, Tier Lists, and AI Skills
Beyond the glossary, GOTBURNOUT provides frontier AI benchmarks, a catalog of over 100,000 AI Agent Skills, production-ready VPS hardening scripts, and an exclusive Pro developer community.