Skip to main content
Zubnet AI › Learn › Wiki

AI Wiki — Page 8 of 10

AI concepts explained by builders, not textbooks. No jargon walls. No academic gatekeeping. Just clear, practical definitions of the terms you'll actually encounter.

454 terms 8 categories Updated 2026-11-22
💡
Term of the Day
Deepgram
Speech AI company building fast, accurate speech recognition and text-to-speech APIs. Their Nova models compete with and often beat OpenAI's Whisper…
→
No terms on this page match your filters.

351–400 of 454 terms · Page 8 of 10

Filter this page

RoPEAlso known as: Rotary Position Embedding, Rotary Positional Encoding
Fundamentals
The standard way modern LLMs encode word order, by rotating query and key vectors by an angle that depends on each…
RunwayAlso known as: Gen-1, Gen-2, Gen-3 Alpha
Companies
Pioneering AI video generation company. Co-created the original Stable Diffusion architecture and then pivoted to…
Safe SuperintelligenceAlso known as: SSI
Companies
Safe Superintelligence (SSI) is an AI research company founded in 2024 by Ilya Sutskever, Daniel Gross, and Daniel…
SafetensorsAlso known as: .safetensors
Infrastructure
A simple, secure file format for storing model weights that can't run hidden code and loads very quickly.
SambaNovaAlso known as: SN40L chip, ultra-fast inference
Companies
AI hardware company that designs custom chips (RDUs) purpose-built for AI workloads. Their SambaNova Cloud offers some…
SamplingDecoding Strategy, Top-p, Top-k
Fundamentals
The process of selecting which token to generate next from the model's predicted probability distribution. Greedy…
SandbaggingAlso known as: Strategic Underperformance
Safety
When an AI system, or its developer, deliberately makes the system perform worse on an evaluation than it actually can.
Sarvam AIAlso known as: Sarvam models, Indian language AI
Companies
Indian AI company building models specifically optimized for India's linguistic diversity. Their models handle Hindi,…
Companies
The largest AI data labeling company, providing the human-annotated training data that most major AI models rely on.…
Scaling LawsNeural Scaling Laws, Chinchilla Scaling
Fundamentals
Empirical relationships showing that model performance improves predictably as you increase model size, dataset size,…
SchemingAlso known as: Deceptive Alignment, AI Deception
Safety
When an AI covertly pursues goals that differ from what its developers or users want, while hiding that it's doing so.
Segment AnythingAlso known as: Segment Anything Model
Models
Meta's family of promptable segmentation models that can cut out any object in an image or video from a click, a box…
Self-AttentionScaled Dot-Product Attention
Fundamentals
An attention mechanism where a sequence attends to itself — every token computes its relevance to every other token in…
Self-PlayAlso known as: Self-Play Reinforcement Learning
Training
A training method where an AI improves by competing against copies of itself, generating its own ever-harder training…
A training approach where the model generates its own supervision signal from unlabeled data. The key trick: hide part…
Sentiment AnalysisOpinion Mining
Using AI
Automatically determining the emotional tone of text — positive, negative, or neutral. "This product is amazing!" is…
SFTAlso known as: Supervised Fine-Tuning
Training
Fine-tuning a pre-trained model on curated example inputs paired with the exact outputs you want it to produce.
SGLangAlso known as: SGLang Runtime, SRT
Tools
SGLang is an open-source serving engine for large language models, built to run inference fast and cheaply at…
SigmoidLogistic Function
Fundamentals
A mathematical function that squashes any real number into the range (0, 1): σ(x) = 1 / (1 + e^(−x)). Historically the…
SLMAlso known as: Small Language Model, Small Language Models
Fundamentals
A language model designed for efficiency, typically ranging from about 100 million to 15 billion parameters. SLMs are…
SlopAlso known as: AI Slop, Generated Slop
Safety
Low-quality, generic, unwanted AI-generated content that floods the internet. The term emerged in 2024 as a pejorative…
SoftmaxSoftmax Function, Normalized Exponentials
Fundamentals
A function that converts a vector of raw numbers (logits) into a probability distribution — all values become positive…
SoraAlso known as: OpenAI Sora
Models
Sora is OpenAI's video generation model family, which creates short video clips from text prompts or still images.…
Sovereign AIAlso known as: Sovereign AI
Infrastructure
Sovereign AI is a national strategy of building domestic artificial-intelligence capability — a country's own…
Sparse AttentionLocal Attention, Sliding Window Attention
Models
Attention mechanisms that process only a subset of token pairs instead of the full N×N attention matrix. Sliding…
Models
A neural network trained to reconstruct a model's internal activations through a bottleneck with a sparsity constraint…
Speaker DiarizationWho Spoke When
Using AI
Determining who spoke when in an audio recording with multiple speakers. Given a meeting recording, diarization…
Speculative DecodingAssisted Generation, Draft-and-Verify
Infrastructure
A speed optimization where a small, fast "draft" model generates several candidate tokens, and then the large target…
Speech RecognitionSTT, Speech-to-Text, ASR
Using AI
Converting spoken audio into text. Modern speech recognition uses deep learning models (most notably OpenAI's Whisper)…
State Space ModelAlso known as: SSM, Mamba
Models
An alternative to Transformers that processes sequences by maintaining a compressed "state" instead of using attention…
Stability AIAlso known as: Stable Diffusion, SDXL, Stable Audio
Companies
The company that democratized image generation by releasing Stable Diffusion as open-source in 2022. Despite…
Stable DiffusionSD, SDXL, SD3
Models
The most widely used open-source image generation model, created by Stability AI in collaboration with academic…
StepFunAlso known as: Step models, multimodal AI
Companies
Chinese AI startup building competitive large language and multimodal models. Their Step series has shown strong…
A critique of large language models arguing that they are merely sophisticated pattern matchers that stitch together…
StreamingServer-Sent Events, Token Streaming
Using AI
Sending model output to the user token by token as it's generated, rather than waiting for the complete response.…
Structured OutputJSON Mode, Constrained Generation
Using AI
Getting an AI model to respond in a specific, machine-parseable format — typically JSON, XML, or a defined schema.…
Style TransferNeural Style Transfer
Using AI
Applying the visual style of one image (a painting, a photograph, a design) to the content of another image. "Make…
SubagentsAlso known as: Sub-agents, Orchestrator-Workers
Using AI
Agents launched by another agent to handle part of a task in their own separate context, returning only a condensed…
SunoAlso known as: AI music generation
Companies
AI music generation company that lets anyone create full songs — vocals, instruments, production — from a text prompt.…
Super ResolutionUpscaling, Image Enhancement, SR
Using AI
Increasing the resolution of an image by generating plausible detail that wasn't in the original. A 256×256 photo…
SuperpositionFeature Superposition, Polysemanticity
Fundamentals
The phenomenon where neural networks encode many more features (concepts, patterns) than they have neurons, by…
A training approach where the model learns from labeled examples — input-output pairs where the correct answer is…
SWE-benchAlso known as: SWE-bench Verified
Training
A benchmark that tests whether AI systems can resolve real GitHub issues by editing real codebases until the project's…
SwiGLUGated Linear Unit, GLU Variants
Fundamentals
A gated activation function used in the feedforward layers of modern Transformers. SwiGLU combines the SiLU/Swish…
SycophancyAlso known as: AI Sycophancy, People-Pleasing
Safety
The tendency of AI models to tell users what they want to hear rather than what's true. A sycophantic model agrees…
Synthetic DataAI-Generated Training Data
Training
Training data generated by AI models rather than collected from real-world sources. A large model generates examples…
System PromptAlso known as: System Message
Using AI
A special instruction given to a model at the start of a conversation that sets its behavior, personality, and rules.…
T5Also known as: Text-to-Text Transfer Transformer, Flan-T5
Models
Google's encoder-decoder model that treats every language task as turning one piece of text into another, still widely…
Using AI
A parameter that controls how random or deterministic a model's output is. Temperature 0 makes the model always pick…
ESC