Author
Divya Sharma
Divya Sharma writes the ReviewRamble AI glossary: 219 terms so far, each defined in plain English with sources.
AI glossary 219
- Abliteration
- Agent framework
- Agent harness
- Agent hooks
- Agent instructions file (AGENTS.md)
- Agent loop
- Agent memory
- Agent orchestration
- Agent permissions
- Agent sandbox
- Agent skills
- Agent2Agent protocol
- Agentic AI
- Agentic RAG
- AI agent
- AI alignment
- AI assistant
- AI avatar
- AI benchmark
- AI code editor
- AI crawler
- AI image upscaling
- AI red teaming
- AI search engine
- AI supply-chain attack
- AI training data
- AI watermarking
- AIME (benchmark)
- API key
- Approximate nearest neighbor search
- ARC-AGI
- Artificial Analysis Intelligence Index
- Attention mechanism
- Autoregressive model
- Base model
- Batch API
- Benchmark contamination
- BM25
- BrowseComp
- Browser agent
- C2PA content credentials
- Chain-of-thought
- Chunking
- Coding agent
- Computer use
- Context compaction
- Context engineering
- Context rot
- Context window
- Continuous batching
- ControlNet
- Cosine similarity
- CPU offloading
- Cross-encoder
- Data exfiltration (AI agents)
- Data residency
- Decision model
- Deep research (AI)
- Deepfake
- Diffusion language model
- Diffusion model
- Direct preference optimization
- Document parsing
- Elo rating (model arenas)
- Embedding
- Embedding dimension
- EU AI Act
- Few-shot prompting
- Fine-tuning
- Foundation model
- Frontier model
- Generative engine optimization
- GGUF
- Golden dataset
- GPQA
- GPU
- GPU cloud
- GraphRAG
- Grounding
- Guardrails (AI)
- Guidance scale (CFG)
- Hallucination (AI)
- HNSW
- Human evaluation
- Human-in-the-loop
- HumanEval
- Humanity's Last Exam
- Hybrid search
- Image-to-image
- Image-to-video
- Inference
- Inference provider
- Inpainting and outpainting
- Instruct model
- Jailbreak (LLM)
- Knowledge base (AI)
- Knowledge cutoff
- Knowledge distillation
- Knowledge graph
- KV cache
- Large language model
- Late interaction (ColBERT)
- Layout analysis
- Leaderboard (AI)
- Lip sync (AI video)
- LLM evaluation
- LLM gateway
- LLM observability
- LLM routing
- LLM-as-a-judge
- llms.txt
- LMArena
- Local LLM
- LoRA
- Matryoshka embeddings
- MCP client
- MCP resources and prompts
- MCP server
- MCP transport (stdio, Streamable HTTP)
- Metadata filtering
- Mixture of experts
- MMLU
- Model card
- Model Context Protocol
- Model fallback
- Model license
- Model parameters
- Model serving
- Model weights
- MTEB
- Multi-agent system
- Multimodal model
- No-code AI app builder
- NPU
- Numeric precision (FP16, BF16, FP8)
- OAuth for AI agents
- ONNX
- Open source AI
- Open-weight model
- OpenAI-compatible API
- Optical character recognition
- OSWorld
- Output tokens
- PagedAttention
- Parameter-efficient fine-tuning
- pass@k
- PII redaction
- Pretraining
- Prompt caching
- Prompt engineering
- Prompt injection
- Quantization
- Query rewriting
- Rate limit (API)
- ReAct (reasoning and acting)
- Reasoning model
- Regression testing (LLM apps)
- Reranking
- Retrieval metrics (recall@k, nDCG, MRR)
- Retrieval-augmented generation
- Reward model
- RLHF
- Safetensors
- Self-hosting (AI models)
- Semantic cache
- Serverless GPU
- Slash command
- Small language model
- SOC 2
- Sparse retrieval
- Speaker diarization
- Spec-driven development
- Speculative decoding
- Speech-to-speech model
- Speech-to-text
- State space model
- Streaming (LLM output)
- Structured output
- Subagent
- Supervised fine-tuning
- SWE-bench
- Synthetic data
- System prompt
- Table extraction
- Temperature (sampling)
- Tensor parallelism
- Terminal-Bench
- Test-time compute
- Text-to-3D
- Text-to-image
- Text-to-music
- Text-to-speech
- Text-to-video
- Time to first token
- Token (LLM)
- Token pricing
- Tokenizer
- Tokens per second
- Tool calling
- Tool poisoning
- Top-p sampling
- TPU
- Transformer (architecture)
- Unified memory
- Vector database
- Vector search
- Vibe coding
- Vision-language model
- Voice agent
- Voice cloning
- VRAM
- Web crawling for AI
- Weight-space programming
- Workflow automation (AI)
- Workflow vs agent
- World model
- Zero data retention
- Zero-shot
- τ-bench