EXPLORE / 06
Terms index
Specialized AI and software-development terms with concise explanations
Entity records
4377 records · 4/110
- 121
Dual-channel Audio
A technique separating different speakers into independent audio tracks, enabling models to precisely learn turn-taking, interru
Website pending - 122
Adaptive Sampling Temperature
Technique dynamically adjusting generation randomness based on prompt difficulty, e.g., raising temperature for hard prompts to
Website pending - 123
Sharpening Tax
The phenomenon where optimized models narrow their output distribution, increasing single-shot confidence but sacrificing explor
Website pending - 124
Parallel Tool Calling
Agents concurrently execute external APIs or database queries while receiving user instructions, rather than serially executing
Website pending - 125
Agent Identity Management
A mechanism that assigns unique identifiers, access tokens, and granular permissions to autonomous AI agents to ensure traceabil
Website pending - 126
Option Reordering Robustness
Measures a model's ability to maintain consistent outputs when input option orders change, assessing reasoning stability.
Website pending - 127
Self-play Reward
A training paradigm where a model generates candidates and self-evaluates them, using internal comparisons as RL signals.
Website pending - 128
NVLink Fabric
NVIDIA's high-speed point-to-point interconnect technology enabling direct data exchange between GPUs, creating a unified high-b
Website pending - 129
Topology-Aware Scheduling
A resource management strategy that optimizes job placement based on physical connectivity structures (e.g., intra-rack high-spe
Website pending - 130
Firmware
Low-level software embedded in hardware devices that controls basic functions, typically stored in non-volatile memory, which is
Website pending - 131
Cost-Plus Pricing
A pricing strategy where a fixed percentage or amount is added to direct costs as profit, often used in long-term infrastructure
Website pending - 132
Context Parallelism
Distributing different segments of long sequences across multiple devices for parallel processing, addressing memory and compute
Website pending - 133
Context Efficiency
The ability to minimize token usage while retaining critical semantic information by streamlining prompts or optimizing retrieva
Website pending - 134
Affine Transformation
A geometric transformation comprising linear mapping and translation; here referring to simplifying complex non-linear gating op
Website pending - 135
DeltaNet
A variant of linear attention introducing forget gates and residual writing mechanisms to enhance dynamic management of historic
Website pending - 136
Parameter Scale
The number of learnable weights in a model, typically determining its capacity, computational cost, and inference speed.
Website pending - 137
Streaming Representation
Feature extraction and encoding mechanisms for continuous input data streams, supporting real-time processing of dynamically cha
Website pending - 138
Omni Intelligence
A capability system involving unified understanding, reasoning, and generation across multiple modalities such as text, image, a
Website pending - 139
Role Homogenization
A phenomenon in multi-agent collaboration where distinct preset roles converge in behavior patterns, leading to single-perspecti
Website pending - 140
4D Spatiotemporal Representation
A data representation method combining 3D spatial geometry with the temporal dimension to capture dynamic physical interaction p
Website pending - 141
GSM8K
Grade School Math 8K, a benchmark dataset commonly used to evaluate LLM mathematical reasoning capabilities.
Website pending - 142
Decode Throughput
The number of tokens generated per second during the decoding phase, directly reflecting inference speed.
Website pending - 143
Factual Hallucination
The model confidently stating non-existent or unsupported facts, statistics, or causal relationships, often stemming from noise
Website pending - 144
Citation Provenance
The process of tracking external links or data sources cited in generative AI responses to verify if they point to original, rel
Website pending - 145
Instruction-Driven Transcription
A technique leveraging LLM instruction-following capabilities to dynamically adjust ASR output format, style, or content filteri
Website pending - 146
Harness Adapter
A trainable middleware layer external to the LLM that encapsulates instructions, skills, memory, and tool configurations, allowi
Website pending - 147
Level of Development 400
A Level of Development in BIM indicating the model contains sufficient geometric and non-geometric information for fabrication a
Website pending - 148
Best-First Search
A graph traversal algorithm that uses a heuristic function to evaluate node priority, expanding the node with the lowest estimat
Website pending - 149
Molecular Simulation
Computational techniques simulating molecular structures, dynamics, and interactions to predict properties of drug candidates an
Website pending - 150
End-to-End Performance
Comprehensive performance measured from raw input to final output across the entire pipeline, rather than evaluating intermediat
Website pending - 151
Hidden Chain-of-Thought
The internal sequence of reasoning steps an LLM takes before generating a final answer, often treated as core intellectual prope
Website pending - 152
Agentic Autonomy
A behavior pattern where AI systems autonomously plan steps and invoke tools to execute tasks based on goals without human inter
Website pending - 153
PCIe Retimer
A signal relay device that regenerates and reshapes PCIe signals over long distances to ensure data integrity in high-speed comm
Website pending - 154
Memory Interconnect Chip
Specialized ICs like DDR5 RCD and MRCD/MDB that enhance server memory bandwidth and stability, serving as critical components fo
Website pending - 155
800G Optical Module
Optical transceiver module supporting 800 Gbps transmission speed, currently a high-performance network component required for A
Website pending - 156
Formal Logic
A reasoning paradigm based on deterministic rules where input-output relationships are explicit, often used to describe traditio
Website pending - 157
Adversarial Defense
Techniques and strategies designed to mitigate the impact of adversarial attacks (e.g., data exfiltration, prompt injection), in
Website pending - 158
Agent Interoperability
The technical capability allowing different AI agent systems or platforms to share context, invoke tools, and reuse capability m
Website pending - 159
Memory Bandwidth
The rate at which data can be read from or stored into memory, critical for inference efficiency in large models.
Website pending - 160
Online Policy Distillation
A technique that extracts knowledge from teacher models or environmental feedback during training to update student model parame
Website pending