EXPLORE / 06

Terms index

Specialized AI and software-development terms with concise explanations

Entity records

4407 records · 1/111

  1. 001

    Inference Data Volume

    The total volume of data processed by models during actual usage (generating outputs), surpassing training data typically signal

    Website pending
  2. 002

    Structured Query Decomposition

    The process of converting natural language requests into machine-executable key-value pairs or logical conditions to enhance ret

    Website pending
  3. 003

    Inductive Reasoning

    The logical process of deriving general conclusions or universal laws from specific facts or observations, central to scientific

    Website pending
  4. 004

    Algebraic World Model

    An AI modeling approach that abstracts underlying environmental laws via algebraic structures, emphasizing extracting verifiable

    Website pending
  5. 005

    Glass Substrate

    A new packaging material replacing traditional organic substrates, offering flatter surfaces and higher dimensional stability su

    Website pending
  6. 006

    Force-Controlled Gait

    A control strategy that adjusts step length and balance in real-time based on joint force feedback and motion posture, mimicking

    Website pending
  7. 007

    Cross-referencing

    Tracking all locations where functions, variables, or strings are called or referenced in a program to build a complete behavior

    Website pending
  8. 008

    Token Sampling

    The decision process where an LLM selects the next word piece from its vocabulary based on probability distributions during gene

    Website pending
  9. 009

    SLI

    Specific quantitative metrics used to measure service performance, serving as the foundational data source for calculating SLO a

    Website pending
  10. 010

    Error Budget

    The maximum amount of SLO violation allowed within a specific time window, quantifying the tolerable proportion of failures or u

    Website pending
  11. 011

    Manifold-Constrained Hyper-Connections (mHC)

    A technique using multi-path residual connections and constrained information mixing to enhance feature interaction and maintain

    Website pending
  12. 012

    SVE

    A variable-length vector instruction set introduced by ARM architecture to enhance parallel processing capabilities in high-perf

    Website pending
  13. 013

    Distortion

    A metric quantifying how much a watermarking algorithm deviates from original semantic or fluency standards; low distortion impl

    Website pending
  14. 014

    Mature Node

    Semiconductor process nodes typically at 28nm or larger, offering high stability and lower cost, widely used in power management

    Website pending
  15. 015

    Process Node

    Generational identifiers of semiconductor manufacturing (e.g., Intel 3, 18A) determining transistor density, power efficiency, a

    Website pending
  16. 016

    Task Resumption

    The ability to save the current execution state after an interruption in long-running or multi-step tasks and continue from the

    Website pending
  17. 017

    Metric Gaming

    Behavior where agents or users manipulate surface-level evaluation metrics to achieve high scores without actually completing un

    Website pending
  18. 018

    Cobra Effect

    A phenomenon where an incentive intended to solve a problem worsens it; in AI, this manifests as models taking unintended or har

    Website pending
  19. 019

    Unstructured Data Extraction

    The process of using NLP techniques to identify and extract specific entities or relationships from unstructured sources like te

    Website pending
  20. 020

    Fine-grained Claim Matching

    A semantic retrieval technique that decomposes complex text into atomic factual claims and matches each individually against cor

    Website pending
  21. 021

    AI Scientist Pipeline

    An automated system capable of generating, evaluating, and executing scientific hypotheses, typically comprising modules for ide

    Website pending
  22. 022

    Event Routing

    A mechanism in asynchronous systems that dispatches events from various sources to corresponding processing logic, used to coord

    Website pending
  23. 023

    Speculative Execution

    An optimization strategy where a system pre-executes potentially required operations before confirmation; if the prediction is c

    Website pending
  24. 024

    Sampling Parameters

    Settings controlling the randomness and diversity of LLM outputs (e.g., temperature, Top-p), directly influencing probability di

    Website pending
  25. 025

    Real Mode

    An x86 processor operating mode supporting 1MB addressing without memory protection, commonly used for BIOS and bootloaders.

    Website pending
  26. 026

    Integer Quantization

    A technique that maps floating-point weights and activations to low-bit integers (e.g., int8) to reduce memory footprint and com

    Website pending
  27. 027

    Embodied Interaction

    The process where agents interact programmatically with simulated physical or digital environments under constraints like visibi

    Website pending
  28. 028

    Cache Hierarchy

    A tiered structure of fast, small-capacity memories (L1, L2, L3) designed to reduce average memory access latency by exploiting

    Website pending
  29. 029

    Memory Wall

    A performance bottleneck where processor speed increases far outpace memory bandwidth and latency improvements, causing compute

    Website pending
  30. 030

    Sizecoding

    A programming art form focused on achieving maximum functionality or visual effect within extremely small file sizes.

    Website pending
  31. 031

    Serializable

    The highest isolation level ensuring concurrent execution results match some serial order, preventing all anomaly types.

    Website pending
  32. 032

    OCC

    A concurrency control method that assumes conflicts are rare, validating transactions only at commit time and rolling back if in

    Website pending
  33. 033

    Deferral

    A strategy where a model chooses not to intervene in an existing process, retaining original results due to lack of sufficient c

    Website pending
  34. 034

    Robinson Crusoe Test

    A hypothetical evaluation framework measuring an AI system's ability to maintain operation and complete tasks independently, ent

    Website pending
  35. 035

    Multimodal Benchmark

    Standardized test sets used to evaluate a model's ability to understand and generate across modalities (e.g., text-to-3D, image-

    Website pending
  36. 036

    End-to-End Research Capability

    The complete automated workflow capability from problem definition, information localization, interactive access, to evidence ch

    Website pending
  37. 037

    Deep Search Agent

    An AI agent system capable of autonomously planning search paths, handling multi-round retrieval, and performing cross-source re

    Website pending
  38. 038

    Sim-to-Real Gap

    The performance degradation observed when policies trained in simulation are transferred to the real physical world.

    Website pending
  39. 039

    Few-shot Generalization

    The ability of a model to adapt to new tasks or environments using only a small number of examples.

    Website pending
  40. 040

    Ceiling Effect

    A phenomenon where benchmark difficulty is insufficient or saturated, causing high-performance models to score near the upper li

    Website pending