EXPLORE / 06

Terms index

Specialized AI and software-development terms with concise explanations

Entity records

4377 records · 23/110

  1. 881

    Hidden HTML Comments

    Using HTML comment tags (<!-- -->) to store data invisible to frontend rendering but readable by crawlers or parsers, often used

    Website pending
  2. 882

    Intent Compilation

    The automated process of translating high-level natural language requirements or design blueprints into low-level executable cod

    Website pending
  3. 883

    Workflow Benchmark

    An evaluation framework based on real business scenarios, focusing on comprehensive performance in continuous, multi-step intera

    Website pending
  4. 884

    Hardware Agnosticism

    The capability of software frameworks to operate independently of specific underlying hardware architectures, aiming to eliminat

    Website pending
  5. 885

    Model Backdoor

    Hidden logic implanted during training or modification that activates only under specific trigger conditions to perform maliciou

    Website pending
  6. 886

    Abliteration

    A technique that removes a model's safety refusal mechanism by subtracting the weight direction associated with saying 'no'.

    Website pending
  7. 887

    Risk Stratification

    A statistical method that categorizes individuals into different risk levels based on characteristics to guide interventions.

    Website pending
  8. 888

    Blue Team Agent

    Automated agents responsible for monitoring anomalous activities, assessing potential threats, and assisting in defensive respon

    Website pending
  9. 889

    Red Team Agent

    Automated agents that simulate attacker behaviors to proactively discover system vulnerabilities and security weaknesses.

    Website pending
  10. 890

    Model Benchmarking

    A systematic method to quantify and compare the performance of different AI models using standardized datasets and metrics.

    Website pending
  11. 891

    Agentic Monitoring

    A technical paradigm using autonomous AI agents to continuously analyze system telemetry and logs, detecting complex or novel an

    Website pending
  12. 892

    Data-Parallel Attention

    Applying data parallelism to attention layers, distributing computations across devices, often combined with Expert Parallelism

    Website pending
  13. 893

    Red-Blue Teaming

    A methodology that evaluates and improves system security by simulating interactions between attackers (red team) and defenders

    Website pending
  14. 894

    Rank-One Model Editing (ROME)

    一种通过更新单个低秩矩阵来精确修改大语言模型特定事实知识的技术,常被用于研究模型内部机制或实施对抗性篡改。

    Website pending
  15. 895

    Software Ecosystem Lock-in

    A situation where users or developers become dependent on a specific vendor's proprietary software stack, tools, or interfaces,

    Website pending
  16. 896

    Hierarchical Composer

    In this specific architecture, refers to the module responsible for parsing complex object structures level by level, determinin

    Website pending
  17. 897

    Data Localization

    A compliance strategy requiring data to be stored and processed within specific jurisdictions.

    Website pending
  18. 898

    Availability Zone

    Isolated data centre clusters within a geographic region designed for high availability and disaster recovery.

    Website pending
  19. 899

    Native Backend

    A computation execution layer optimized directly for specific hardware architectures, enabling efficient operator scheduling wit

    Website pending
  20. 900

    Expert Parallelism

    A strategy distributing different expert sub-networks of an MoE model across multiple devices for parallel execution to overcome

    Website pending
  21. 901

    Pairwise Comparison

    An evaluation method where raters compare two options rather than scoring them individually, reducing subjective bias and improv

    Website pending
  22. 902

    Fault-Tolerant Quantum Computing

    The capability of quantum computers to reliably execute algorithms despite noise and errors, achieved through techniques like qu

    Website pending
  23. 903

    Dataflow Architecture

    A computational model where instruction execution is triggered by data availability rather than strict program counter order, of

    Website pending
  24. 904

    Penetration Testing

    An authorized testing method that evaluates computer system security by simulating real-world attacks.

    Website pending
  25. 905

    Dynamic Highlights

    A technique that automatically determines how much text to return from each page within a shared budget, optimizing search resul

    Website pending
  26. 906

    Tool Hallucination

    The phenomenon where LLMs incorrectly generate non-existent parameters, function names, or execution logic when calling external

    Website pending
  27. 907

    Keyframe Interpolation

    A technique using sparse keyframe images as conditional constraints to guide the model in generating intermediate frames for mot

    Website pending
  28. 908

    Diffusion LLM

    Large-scale pre-trained models applying diffusion principles to NLP, aiming to enhance generation parallelism and speed.

    Website pending
  29. 909

    Chained Tool Calls

    An agent sequentially invoking multiple external tools or APIs in a logical order within a single task, where the output of one

    Website pending
  30. 910

    Cross-Verification

    The process where an agent actively verifies the truthfulness of information or claims provided by another agent to prevent erro

    Website pending
  31. 911

    Multimodal Prompting

    A technique combining text, image, and other modalities as inputs to guide models in generating specific outputs or performing c

    Website pending
  32. 912

    Video Language Model

    A multimodal foundation model designed to understand temporal information in videos and map it to text semantic spaces.

    Website pending
  33. 913

    Teleoperation Data Collection

    A method recording human operator actions via remote control or wearable devices to provide high-quality demonstration data for

    Website pending
  34. 914

    Multi-turn Dialogue Evaluation

    A systematic approach evaluating context consistency, intent retention, and long-term task completion quality in continuous inte

    Website pending
  35. 915

    Controlled Experimentation

    A method to evaluate improvements by generating reference trajectories and comparing outputs from different strategies when expl

    Website pending
  36. 916

    Effective Cluster Compute

    The actual usable computational power of a large-scale cluster after accounting for interconnect efficiency, scheduling overhead

    Website pending
  37. 917

    Interconnect Chip

    High-speed hardware components connecting compute nodes, storage, and networks, determining internal cluster data transmission b

    Website pending
  38. 918

    Multi-scale Modeling

    In bioinformatics, refers to computational methods handling data interactions across different hierarchical levels, from molecul

    Website pending
  39. 919

    Automotive Safety Integrity Level

    Safety levels defined in ISO 26262 (A-D) to assess risk of failure in electronic/electrical systems, with D being the most strin

    Website pending
  40. 920

    Pipelining

    A technique where multiple commands are sent to the server at once without waiting for intermediate results, reducing network ro

    Website pending