Clear definitions for the era of Agentic AI and Spatial Intelligence.
20 terms starting with "V"
Specialized database for high-dimensional vectors (embeddings). Enables semantic similarity search instead of exact keyword matches. Core infrastructure for RAG, recommendation systems, and multimodal search. Leading solutions: Pinecone, Weaviate, Qdrant, Chroma, pgvector.
Vector Databases are specialized databases designed to store and efficiently query high-dimensional vector embeddings, which represent the semantic meaning of data. These databases are essential for applications like semantic search, recommendation systems, and retrieval-augmented generation.
A form of private equity financing provided to early-stage, high-growth startups in exchange for equity. In the AI sector, VC funding has become critical for scaling compute-intensive AI companies, with global AI VC investment exceeding $50B annually.
A training-free prompting strategy to overcome Mode Collapse. The model is asked to verbalize an explicit probability distribution over multiple possible answers and then sample from it. Increases output diversity by 1.6-2.1× without quality loss.
A term coined by Andrej Karpathy in February 2025 for AI-assisted software development where developers focus on the vision and AI writes the code. Collins Word of the Year 2025.
AI-assisted software development where you describe the desired outcome in natural language and AI generates the code.
AI-assisted software development using natural language.
A development approach where programmers rely heavily on AI to generate code from natural language, accepting output with minimal review — prioritizing speed over understanding and quality.
AI-assisted software development where you describe the desired outcome in natural language.
Vision-Language Models (VLMs) are AI models that combine computer vision and natural language processing to understand and reason about images and text simultaneously. They can perform tasks such as image captioning, visual question answering, and cross-modal retrieval.
A performance metric for language models measuring how efficiently a model compresses validation data. Calculated as bits per byte of text, lower values indicate better compression and thus better model performance. Used as the optimization target in Karpathy autoresearch because it is automatically computable, requiring no human judgment between iterations.
Google's latest video generation AI model that powers Google Flow, offering high-quality video synthesis with support for complex transitions and native audio generation.
Google latest video generation AI model that powers Google Flow offering high-quality video synthesis with support for complex transitions and native audio generation.
A technique to combat mode collapse by explicitly instructing the AI model through natural language prompts to generate diverse outputs rather than relying on temperature or sampling parameters.
In the context of AI agents, a vertical connection refers to the link between an agent and the external tools, databases, and APIs it uses to perform tasks.
The connection between an AI agent and the tools, databases, and APIs it needs to access external information and perform tasks. In the AI protocol landscape, MCP handles this vertical dimension.
A development approach where AI generates code from high-level human guidance.
The negative consequences after extensive vibe coding: accumulated technical debt, unmaintainable code, security vulnerabilities, and the realization that AI-generated code requires engineering rigor.
The automated process of identifying security weaknesses and potential vulnerabilities in software, networks, or systems. Modern approaches range from rule-based static analysis to AI-powered contextual code reasoning.
Vendor lock-in in AI refers to the dependency on a single AI provider's models, tools, and ecosystem, making it costly to switch. GitHub's Agent HQ addresses this by supporting multiple AI agents (Claude, Codex, Copilot), allowing developers to avoid single-vendor dependency and choose the best tool for each task.