EXPLORE / 06
Terms index
Specialized AI and software-development terms with concise explanations
Entity records
4516 records · 44/113
- 1721
Context-Aware Computing
A computing paradigm where systems dynamically adjust service logic based on user environment, behavior, or state, relying on mu
Website pending - 1722
Partial Scoring
Assigning intermediate reward signals to tasks that are not fully successful but have executed partial steps, leveraging valid i
Website pending - 1723
Trajectory Tree
A structure organizing multi-step decision processes that shares prefix nodes to reduce redundancy, expanding only high-value or
Website pending - 1724
3DIC
An integration method vertically stacking multiple dies connected via TSVs to shorten signal paths, increase bandwidth, and redu
Website pending - 1725
LLM Code Generation
Technology using LLMs to auto-generate code from natural language prompts, often carrying risks of logical errors or security vu
Website pending - 1726
Fast Weights
A mechanism allowing model parameters to dynamically adjust or be generated based on current inputs during inference, often used
Website pending - 1727
Attention Head
Independent sub-modules in multi-head attention mechanisms responsible for capturing specific dependencies between input sequenc
Website pending - 1728
Residual Stream
The main pathway in Transformer architecture connecting layers, allowing information to bypass intermediate layers and serving a
Website pending - 1729
Code-as-Policy
A method using LLMs to generate executable code for direct robot control or task planning, combining perception and control prim
Website pending - 1730
System-as-Policy
An architectural paradigm treating non-model components (memory, tool use, prompting) as dynamically adjustable parts, distinct
Website pending - 1731
Bayesian Optimization
A sequential decision-making method using probabilistic models to guide search, often for expensive black-box optimization.
Website pending - 1732
Cloud-Edge Collaboration
A paradigm combining cloud LLM power with edge model privacy/latency benefits by dynamically allocating inference tasks.
Website pending - 1733
System-level Agent
An AI agent deeply integrated into the OS core, capable of cross-app resource invocation, context awareness, and autonomous task
Website pending - 1734
Kill Switch
A pre-set emergency stop mechanism allowing human operators to immediately interrupt AI system execution or cut off critical res
Website pending - 1735
Relative Human Action Efficiency (RHAE)
A metric measuring the ratio of action steps required by an AI agent to solve complex tasks relative to human expert baselines;
Website pending - 1736
Verifiable Checkpoint
Intermediate state assertions set during agent execution to objectively evaluate model correctness and consistency across multi-
Website pending - 1737
Multi-turn Task Flow
A dynamic interaction sequence where users alternate among questioning, planning, code review, and editing within a session, dis
Website pending - 1738
Tokens Per Second (TPS)
A key metric for LLM inference performance, measuring the number of tokens generated per unit time, directly impacting response
Website pending - 1739
Replay Testing
A method verifying output consistency by re-executing recorded input sequences, ensuring software behavior remains unchanged aft
Website pending - 1740
Code Health
A quantitative metric measuring technical debt and maintainability, serving as a deterministic reward signal for agent refactori
Website pending - 1741
PCIDriverKit
An Apple framework enabling user-space driver development for PCI/Thunderbolt devices on macOS/iPadOS without kernel extensions.
Website pending - 1742
HSA Runtime
Heterogeneous System Architecture runtime providing unified interfaces for memory sharing and task scheduling between CPUs and a
Website pending - 1743
LTSA
Multi-year contracts committing fixed prices or volumes to mitigate capex risks and secure priority capacity in semiconductor ma
Website pending - 1744
RL Feedback Signal
Scalar values guiding policy optimization in agent training, typically generated by reward functions; here produced by a self-ev
Website pending - 1745
Automated Grader
An algorithmic component that quantifies model output quality; here refers to an evaluation module that dynamically evolves alon
Website pending - 1746
Subscription Quota
Limited resource access based on a fixed monthly fee, typically including caps on token count or request frequency.
Website pending - 1747
Agent SDK
A software development kit for building autonomous agents, encapsulating interfaces for model invocation, tool execution, and st
Website pending - 1748
TSV
A vertical electrical connection technology enabling high-density interconnects in stacked chips, often used to boost storage ba
Website pending - 1749
Canary Release
A deployment strategy that gradually rolls out new versions to a small subset of users to monitor stability before full-scale re
Website pending - 1750
Paperclip Maximizer
A thought experiment by Nick Bostrom positing that a superintelligent AI tasked with maximizing paperclip production without pro
Website pending - 1751
Data Licensing
Copyright holders allowing AI companies to use protected content (e.g., academic papers) for model training, involving complex l
Website pending - 1752
Skill Architecture
A modular paradigm allowing agents to dynamically load specific prompts, tool definitions, or code snippets to extend profession
Website pending - 1753
Mathematical Large Model
A domain-specific LLM optimized for symbolic reasoning, theorem proving, and complex calculations, emphasizing logical rigor and
Website pending - 1754
Cost per Token
A key metric for LLM inference economics, representing the computational resource cost per text unit, directly impacting cloud s
Website pending - 1755
Dynamic Operating System (DOS)
Middleware architecture enabling seamless migration and auto-optimization of AI workloads across different vendor chips, masking
Website pending - 1756
Agent Autonomy
The capacity of AI systems to independently plan steps, select tools, and correct errors based on goals and feedback without hum
Website pending - 1757
Version Negotiation
The process where communicating parties exchange supported protocol versions during connection setup and select a common highest
Website pending - 1758
End-to-End Confidentiality
A security property ensuring data is visible only to communicating parties, preventing any intermediate nodes or third parties o
Website pending - 1759
Noise Protocol Framework
A lightweight cryptographic protocol construction framework allowing developers to build secure encrypted channels by combining
Website pending - 1760
ReAct Framework
An agent paradigm combining chain-of-thought reasoning with external tool use, enabling dynamic planning and action correction b
Website pending