The AI terms that
actually matter.
Clear, technical definitions of 107 key concepts — written by AI engineers, not marketers.
Retrieval-Augmented Generation (RAG)
RAG is technique that combines information retrieval from external sources with text generation, resulting in factually accurate and context-aware AI.
Read definitionRotary Positional Embedding (RoPE)
Rotary Positional Embedding (RoPE) is the positional encoding used by nearly every modern open-weight LLM: it rotates query and key vectors by an angle proportional to their sequence position, so the attention score between two tokens depends only on their relative distance, the property that makes post-training context-length extension (YaRN, NTK-aware scaling, LongRoPE) possible.
Read definitionRRF (Reciprocal Rank Fusion)
Reciprocal Rank Fusion is a simple, training-free algorithm for combining multiple ranked result lists — typically dense vector search and sparse keyword search — into a single, more accurate ranking, using only each item's rank position rather than its raw score.
Read definitionSeedance 2.0
Glossary about Seedance 2.0
Read definitionSeedream 5.0 Lite
Seedream 5.0 Lite is a multimodal image generation model on BytePlus ModelArk, supporting text and image inputs for high-quality, instruction-aligned image generation and editing workflows.
Read definitionSemantic Cache
Semantic cache is a caching technique where LLM responses are stored and retrieved based on semantic similarity between queries rather than exact string matching — dramatically reducing redundant LLM calls and API costs when users ask questions that mean the same thing in different words.
Read definitionSemantic Chunking
Semantic chunking splits a document into retrieval chunks at points where consecutive sentences stop being topically similar, using embedding-similarity drops or discourse boundaries instead of a fixed character or token count. It trades extra compute at ingest time for chunks that hold together as coherent units of meaning.
Read definitionSliding Window Attention (SWA)
Sliding Window Attention (SWA) restricts each token to attending only within a fixed-size local neighborhood instead of the full sequence, cutting attention cost from quadratic to linear in sequence length; stacked across layers its effective receptive field still reaches far, and interleaving it with periodic full-attention layers (Mistral, Gemma, GPT-OSS) is the standard recipe modern LLMs use to control KV cache size at long context.
Read definitionSora AI Model
Sora is OpenAI's groundbreaking text-to-video AI model and can generate high-def videos of up to 1 min duration.
Read definitionSovereign AI
Sovereign AI refers to the strategic development, deployment, and control of artificial intelligence capabilities by a nation, region, or organization to ensure data privacy, security, and cultural alignment, independent of foreign or third-party infrastructure.
Read definitionSpeaker Diarization Models
AI models designed to partition an audio stream into homogeneous segments according to the speaker identity, effectively answering the question 'who spoke when'.
Read definitionSpeculative Decoding
Speculative decoding accelerates LLM inference by having a small, cheap draft model propose several tokens ahead, then verifying all of them in a single forward pass of the large target model. Accepted tokens are emitted for free, and a rejection sampling step guarantees the output distribution is mathematically identical to normal decoding, so the speedup costs nothing in quality.
Read definitionStay ahead of the curve
Weekly newsletter on agentic AI, LLMs, and what we're building at Superteams — straight to your inbox.
Ready to ship AI in production?
We deploy fractional AI teams that deliver production-grade systems in 30–90 days. No fluff, no obligation.