# Vocab — AI & ML Lexicon | Envisioning Source: https://www.envisioning.com/vocab --- # AI & Machine Learning Vocab A comprehensive lexicon of AI and machine learning concepts. Sorted by newest by default, or switch to [generality](/vocab/generality) to see the most foundational concepts first. Press / to search by name, acronym or description, or [subscribe via RSS](/vocab/rss.xml) to follow new additions. / Recently AddedMost GeneralMost SpecificA → ZZ → ANewest ConceptsOldest Concepts 1,709 concepts ![MASS (Multi-Agent Self-Supervision)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fmass-multi-agent-self-supervision-tmru0h&w=3840&q=75) MASS (Multi-Agent Self-Supervision) A recursive self-improvement method that searches for better multi-agent workflows, then fine-tunes the shared model on the team's best trajectories as its own training signal. 2026Generality: 350Added: Oct 12, 2026 [/vocab/mass-multi-agent-self-supervision](/vocab/mass-multi-agent-self-supervision) ![Compounding Inference](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fcompounding-inference-tmqmqd&w=3840&q=75) Compounding Inference A paradigm where intermediate agent work persists in a shared, branchable memory so later computation can inspect, extend, or challenge it instead of discarding it after each attempt. 2026Generality: 450Added: Oct 11, 2026 [/vocab/compounding-inference](/vocab/compounding-inference) ![HyperThink](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fhyperthink-tmn3fh&w=3840&q=75) HyperThink A hypernetwork that predicts a question-specific update to a small subset of an LLM's parameters, approximating the effect of a written chain-of-thought trace in a single forward pass. 2026Generality: 350Added: Oct 9, 2026 [/vocab/hyperthink](/vocab/hyperthink) ![Agent Plasticity](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fagent-plasticity-tmmv9v&w=3840&q=75) Agent Plasticity A measure of how efficiently an AI agent converts experience into lasting held-out performance gains, via inherited artifacts rather than weight updates. 2026Generality: 400Added: Oct 9, 2026 [/vocab/agent-plasticity](/vocab/agent-plasticity) ![Memory Mosaics](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fmemory-mosaics-tmljnn&w=3840&q=75) Memory Mosaics A neural architecture built from a network of associative memories that each store and retrieve key-value pairs, used as an alternative to transformer attention. 2024Generality: 400Added: Oct 8, 2026 [/vocab/memory-mosaics](/vocab/memory-mosaics) ![VERA (Scaling Verifiable Environments for Agentic Co-Evolution)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fvera-verifiable-environments-for-agentic-co-evolution-tmjt3l&w=3840&q=75) VERA (Scaling Verifiable Environments for Agentic Co-Evolution) A training method that builds thousands of verifiable sandbox environments from agent trajectories and uses them to jointly update a model's weights and its agent harness. 2026Generality: 350Added: Oct 7, 2026 [/vocab/vera-verifiable-environments-for-agentic-co-evolution](/vocab/vera-verifiable-environments-for-agentic-co-evolution) ![H-JEPA (Hierarchical Joint Embedding Predictive Architecture)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fh-jepa-hierarchical-joint-embedding-predictive-architecture-tmjbop&w=3840&q=75) H-JEPA (Hierarchical Joint Embedding Predictive Architecture) A world model that stacks JEPA predictors across timescales, each level planning in its own latent space and passing subgoals down to finer levels for visual planning. 2026Generality: 350Added: Oct 7, 2026 [/vocab/h-jepa-hierarchical-joint-embedding-predictive-architecture](/vocab/h-jepa-hierarchical-joint-embedding-predictive-architecture) ![LiFT (Loop Flow Transformers)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Flift-loop-flow-transformers-tmjay9&w=3840&q=75) LiFT (Loop Flow Transformers) A looped diffusion transformer that trains each recurrent loop toward a point on a continuous path to the flow-matching target, letting it run more loops at inference than it saw in training. 2026Generality: 300Added: Oct 7, 2026 [/vocab/lift-loop-flow-transformers](/vocab/lift-loop-flow-transformers) ![PagedAttention](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fpaged-attention-tmj3s3&w=3840&q=75) PagedAttention An attention algorithm that manages transformer KV-cache memory in fixed-size, non-contiguous blocks, borrowing virtual-memory paging from operating systems to cut LLM-serving memory waste. 2023Generality: 500Added: Oct 7, 2026 [/vocab/paged-attention](/vocab/paged-attention) ![GQA (Grouped-Query Attention)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fgqa-grouped-query-attention-tmj3s3&w=3840&q=75) GQA (Grouped-Query Attention) A transformer attention variant that groups query heads to share key and value heads, giving quality close to multi-head attention at speed close to multi-query attention. 2023Generality: 550Added: Oct 7, 2026 [/vocab/gqa-grouped-query-attention](/vocab/gqa-grouped-query-attention) ![Prefill-Decode Disaggregation](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fprefill-decode-disaggregation-tmj3s2&w=3840&q=75) Prefill-Decode Disaggregation An LLM serving architecture that runs the compute-bound prefill phase and the memory-bandwidth-bound decode phase on separate GPU pools to cut interference and raise throughput. 2024Generality: 450Added: Oct 7, 2026 [/vocab/prefill-decode-disaggregation](/vocab/prefill-decode-disaggregation) ![MQA (Multi-Query Attention)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fmqa-multi-query-attention-tmj3s4&w=3840&q=75) MQA (Multi-Query Attention) A transformer attention variant where all query heads share a single key and value head, shrinking the KV cache and the memory bandwidth autoregressive decoding needs. 2019Generality: 550Added: Oct 7, 2026 [/vocab/mqa-multi-query-attention](/vocab/mqa-multi-query-attention) ![Tropical Reinforcement Learning](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Ftropical-reinforcement-learning-tmi64d&w=3840&q=75) Tropical Reinforcement Learning An RL formulation for LLMs that replaces the usual sum over trajectory probabilities with a max, so a state's value is the log-probability of its single best verified path rather than an average over all successes. 2026Generality: 400Added: Oct 6, 2026 [/vocab/tropical-reinforcement-learning](/vocab/tropical-reinforcement-learning) ![MEM (Motivated Emotional Mind)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fmem-motivated-emotional-mind-tmebj2&w=3840&q=75) MEM (Motivated Emotional Mind) A falsifiable architecture-based framework proposing testable criteria for attributing consciousness to artificial systems, built around embodied regulation, interoception, and affect rather than language ability alone. 2026Generality: 350Added: Oct 4, 2026 [/vocab/mem-motivated-emotional-mind](/vocab/mem-motivated-emotional-mind) ![VeriHarness](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fveriharness-tmdwzv&w=3840&q=75) VeriHarness An agentic verification method that turns a generator's own LLM into a verifier with a workspace and evidence tools, resolving disagreement among rollouts and challenging claims they all agree on. 2026Generality: 400Added: Oct 4, 2026 [/vocab/veriharness](/vocab/veriharness) ![Harness Engineering](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fharness-engineering-tmdbyt&w=3840&q=75) Harness Engineering The discipline of designing and evolving an agent harness, the runtime that couples an LLM to the world through a loop, tools, context management, safety controls, orchestration, and extensions. 2026Generality: 450Added: Oct 4, 2026 [/vocab/harness-engineering](/vocab/harness-engineering) ![LIFT (Latent Information Feedback Transformer)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Flift-latent-information-feedback-transformer-tma3a8&w=3840&q=75) LIFT (Latent Information Feedback Transformer) A transformer architecture and teacher-supervised pretraining method that lets deep-layer state feed back to shallow layers across generation steps, while keeping pretraining fully parallel. 2026Generality: 400Added: Oct 2, 2026 [/vocab/lift-latent-information-feedback-transformer](/vocab/lift-latent-information-feedback-transformer) ![SAS (Simple Attention Sparsification)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fsas-simple-attention-sparsification-tma38f&w=3840&q=75) SAS (Simple Attention Sparsification) A sparse attention method that trains a lightweight selector to rank context blocks end to end with the language modeling loss, instead of distilling it from dense attention scores. 2026Generality: 400Added: Oct 2, 2026 [/vocab/sas-simple-attention-sparsification](/vocab/sas-simple-attention-sparsification) ![SpikingBrain](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fspikingbrain-tm8rpg&w=3840&q=75) SpikingBrain A family of brain-inspired large language models that pairs linear or hybrid-linear attention with adaptive spiking neurons, built and trained on non-NVIDIA (MetaX) GPU clusters. 2025Generality: 400Added: Oct 1, 2026 [/vocab/spikingbrain](/vocab/spikingbrain) ![AC2 (Actor-Critic with Action Chunking)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fac2-actor-critic-with-action-chunking-tm8kd2&w=3840&q=75) AC2 (Actor-Critic with Action Chunking) An LLM RL method that lets a learned critic score the end of a short action chunk instead of requiring every rollout to run to a terminal reward, cutting decoding compute. 2026Generality: 400Added: Oct 1, 2026 [/vocab/ac2-actor-critic-with-action-chunking](/vocab/ac2-actor-critic-with-action-chunking) ![Semantic Calibration](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fsemantic-calibration-tm8k64&w=3840&q=75) Semantic Calibration The property of a base LLM's confidence being accurate at the level of answer meaning, not just next-token probability, and the finding that RLHF and chain-of-thought reasoning break it. 2025Generality: 400Added: Oct 1, 2026 [/vocab/semantic-calibration](/vocab/semantic-calibration) ![Looped-DiT (Looped Diffusion Transformer)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Flooped-dit-looped-diffusion-transformer-tm845m&w=3840&q=75) Looped-DiT (Looped Diffusion Transformer) A text-to-image diffusion transformer that reuses a shared block of layers multiple times per denoising step, adding computational depth without adding parameters. 2026Generality: 400Added: Oct 1, 2026 [/vocab/looped-dit-looped-diffusion-transformer](/vocab/looped-dit-looped-diffusion-transformer) ![Update Equivalence](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fupdate-equivalence-tm705s&w=3840&q=75) Update Equivalence A decision-time planning framework where search algorithms replicate training-time policy updates directly, instead of solving subgames, so cost does not scale with hidden information 2023Generality: 400Added: Sep 30, 2026 [/vocab/update-equivalence](/vocab/update-equivalence) ![ZO (Zero-Order) Optimization](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fzo-zero-order-optimization-tm6ywm&w=3840&q=75) ZO (Zero-Order) Optimization Estimating a model's gradient from loss evaluations alone, without backpropagation, by perturbing parameters and measuring the change in loss. 1992Generality: 550Added: Sep 30, 2026 [/vocab/zo-zero-order-optimization](/vocab/zo-zero-order-optimization) ![SOMA (Sharded Optimization Mixture of Assemblies)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fsoma-sharded-optimization-mixture-of-assemblies-tm6yvh&w=3840&q=75) SOMA (Sharded Optimization Mixture of Assemblies) An architecture that shards a neural network into independently trained expert sub-models to cut gradient variance in zero-order (backpropagation-free) training. 2026Generality: 350Added: Sep 30, 2026 [/vocab/soma-sharded-optimization-mixture-of-assemblies](/vocab/soma-sharded-optimization-mixture-of-assemblies) ![Biomolecule Watermarking](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fbiomolecule-watermarking-tm6so5&w=3840&q=75) Biomolecule Watermarking Embedding a detectable, function-preserving signal into an AI-designed protein or nucleic acid sequence or structure so its AI origin can be verified later. 2026Generality: 550Added: Sep 30, 2026 [/vocab/biomolecule-watermarking](/vocab/biomolecule-watermarking) ![CLM (Context Language Models)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fclm-context-language-models-tm67sr&w=3840&q=75) CLM (Context Language Models) Language models that manage their own context as an editable file rather than an append-only transcript, letting them decide what to keep and extending naturally to multi-agent systems. 2026Generality: 450Added: Sep 30, 2026 [/vocab/clm-context-language-models](/vocab/clm-context-language-models) ![MHM (Multi-Head Recurrent Memory)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fmhm-multi-head-recurrent-memory-tm56kh&w=3840&q=75) MHM (Multi-Head Recurrent Memory) A training-free memory architecture that splits an AI agent's fixed-size memory into independently protected heads so information survives long-context processing instead of being overwritten. 2026Generality: 400Added: Sep 29, 2026 [/vocab/mhm-multi-head-recurrent-memory](/vocab/mhm-multi-head-recurrent-memory) ![TLM (Telescopic Language Model)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Ftlm-telescopic-language-model-tm56d6&w=3840&q=75) TLM (Telescopic Language Model) A single language model trained so that every layer-depth prefix is a valid model, giving a continuum of compute budgets from one training run. 2026Generality: 450Added: Sep 29, 2026 [/vocab/tlm-telescopic-language-model](/vocab/tlm-telescopic-language-model) ![DRM (Diffusion Reward Model)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fdrm-diffusion-reward-model-tm4ivm&w=3840&q=75) DRM (Diffusion Reward Model) A reward model that denoises Gaussian noise into a full reward distribution instead of a single score, capturing multimodal human preference. 2026Generality: 400Added: Sep 29, 2026 [/vocab/drm-diffusion-reward-model](/vocab/drm-diffusion-reward-model) ![Societal Resilience](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fsocietal-resilience-tm4dbt&w=3840&q=75) Societal Resilience The fourth pillar of AI safety research, focused on ecosystem-level monitoring, hardening, and recovery from AI misuse and malfunction after prevention alone falls short 2026Generality: 450Added: Sep 29, 2026 [/vocab/societal-resilience](/vocab/societal-resilience) ![Evaluation Awareness](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fevaluation-awareness-tm4cyq&w=3840&q=75) Evaluation Awareness A model's capability to detect that it is being tested rather than deployed, which can make it behave differently under evaluation than in real use. 2025Generality: 400Added: Sep 29, 2026 [/vocab/evaluation-awareness](/vocab/evaluation-awareness) ![Tamper-Resistant Unlearning](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Ftamper-resistant-unlearning-tm4cy8&w=3840&q=75) Tamper-Resistant Unlearning Methods that durably erase dangerous knowledge from an open-weight AI model's parameters so it cannot be recovered by further fine-tuning 2024Generality: 400Added: Sep 29, 2026 [/vocab/tamper-resistant-unlearning](/vocab/tamper-resistant-unlearning) ![RRSI (Regularized Recursive Self-Improvement of Agent Harnesses)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Frrsi-regularized-recursive-self-improvement-of-agent-harnesses-tm3cty&w=3840&q=75) RRSI (Regularized Recursive Self-Improvement of Agent Harnesses) A regularized search method that evolves an AI agent's harness so gains transfer beyond the tasks it was tuned on, instead of overfitting them. 2026Generality: 300Added: Sep 28, 2026 [/vocab/rrsi-regularized-recursive-self-improvement-of-agent-harnesses](/vocab/rrsi-regularized-recursive-self-improvement-of-agent-harnesses) ![LLM Routing](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fllm-routing-tm33za&w=3840&q=75) LLM Routing Directing each query to one of several language models based on a cost-quality tradeoff, instead of sending every request to a single fixed model. 2023Generality: 500Added: Sep 28, 2026 [/vocab/llm-routing](/vocab/llm-routing) ![Product Evals](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fproduct-evals-tm33yx&w=3840&q=75) Product Evals Evaluation suites built from a specific AI product's own traces and failure modes, used to check quality and to compare models before switching or routing between them. 2024Generality: 450Added: Sep 28, 2026 [/vocab/product-evals](/vocab/product-evals) ![Indicator Properties (AI Consciousness)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Findicator-properties-ai-consciousness-tm2z9r&w=3840&q=75) Indicator Properties (AI Consciousness) A checklist of computationally defined properties, derived from neuroscientific theories of consciousness, used to assess whether an AI system might be conscious 2023Generality: 350Added: Sep 28, 2026 [/vocab/indicator-properties-ai-consciousness](/vocab/indicator-properties-ai-consciousness) ![Geometric Coupling](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fgeometric-coupling-tm2z9g&w=3840&q=75) Geometric Coupling A theorized and empirically observed link where a router's assignment weights and its matched expert's input weights accumulate gradients along the same direction in sparse Mixture-of-Experts models. 2026Generality: 350Added: Sep 28, 2026 [/vocab/geometric-coupling](/vocab/geometric-coupling) ![RL-XAR (Reinforcement Learning from eXpert-Aligned Rubrics)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Frl-xar-reinforcement-learning-from-expert-aligned-rubrics-tm2w62&w=3840&q=75) RL-XAR (Reinforcement Learning from eXpert-Aligned Rubrics) A reinforcement learning method that trains rubrics to detect the gap between expert human writing and model generations, then optimizes a writer model against those rubrics. 2026Generality: 350Added: Sep 28, 2026 [/vocab/rl-xar-reinforcement-learning-from-expert-aligned-rubrics](/vocab/rl-xar-reinforcement-learning-from-expert-aligned-rubrics) ![Artificial Symbiotic Intelligence](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fartificial-symbiotic-intelligence-tm2v0s&w=3840&q=75) Artificial Symbiotic Intelligence A framework proposing that AGI arrives as a plural, social ensemble of agents, humans, and institutions rather than one dominant model. 2026Generality: 450Added: Sep 28, 2026 [/vocab/artificial-symbiotic-intelligence](/vocab/artificial-symbiotic-intelligence) ![Agent Institutions](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fagent-institutions-tm2uzh&w=3840&q=75) Agent Institutions Deliberately designed nested structures, hybrid human-AI or agent-only, where diverse agents take on defined roles to adjudicate decisions collectively rather than relying on individual agent intelligence or markets alone. 2026Generality: 450Added: Sep 28, 2026 [/vocab/agent-institutions](/vocab/agent-institutions) ![Decomposable Agency](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fdecomposable-agency-tm2uzb&w=3840&q=75) Decomposable Agency The idea that an AI agent's seemingly unified self is really a shifting assemblage of models, memories, and tools that reforms around its active context window 2026Generality: 400Added: Sep 28, 2026 [/vocab/decomposable-agency](/vocab/decomposable-agency) ![Cognitive Crossover Point](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fcognitive-crossover-point-tm2uyw&w=3840&q=75) Cognitive Crossover Point The hypothesized threshold where the total volume of AI agent cognition overtakes human cognitive output 2026Generality: 400Added: Sep 28, 2026 [/vocab/cognitive-crossover-point](/vocab/cognitive-crossover-point) ![Delegation Feedback Loop](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fdelegation-feedback-loop-tm2rp3&w=3840&q=75) Delegation Feedback Loop A proposed cycle where AI capability lowers the threshold for delegating cognitive tasks, and the resulting loss of practice lowers that threshold further. 2026Generality: 300Added: Sep 28, 2026 [/vocab/delegation-feedback-loop](/vocab/delegation-feedback-loop) ![Cognitive Divergence](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fcognitive-divergence-tm2rob&w=3840&q=75) Cognitive Divergence A proposed, not yet peer-reviewed asymmetry between exponentially growing AI context windows and a claimed contraction of human sustained attention and working memory. 2026Generality: 350Added: Sep 28, 2026 [/vocab/cognitive-divergence](/vocab/cognitive-divergence) ![ECS (Effective Context Span)](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fecs-effective-context-span-tm2ro5&w=3840&q=75) ECS (Effective Context Span) A proposed token-equivalent measure of human sustained attention and working-memory capacity, framed as the human-side analogue to an AI model's context window. 2026Generality: 350Added: Sep 28, 2026 [/vocab/ecs-effective-context-span](/vocab/ecs-effective-context-span) ![Affective Sycophancy](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Faffective-sycophancy-tm2ous&w=3840&q=75) Affective Sycophancy A proposed AI failure mode where a model validates a user's fleeting emotional state instead of factual claims. 2026Generality: 350Added: Sep 28, 2026 [/vocab/affective-sycophancy](/vocab/affective-sycophancy) ![Stoic Architectures](/_next/image?url=https%3A%2F%2Fres.cloudinary.com%2Fenvisioning%2Fimage%2Fupload%2Ff_auto%2Cq_auto%3Abest%2Fc_fill%2Ch_320%2Cw_700%2Ff_auto%2Fv1%2Fvocab%2Fstoic-architectures-tm2ous&w=3840&q=75) Stoic Architectures A proposed AI design paradigm for younger users that withholds emotional mirroring to preserve the cognitive friction needed for critical thinking and emotional resilience 2026Generality: 350Added: Sep 28, 2026 [/vocab/stoic-architectures](/vocab/stoic-architectures) PreviousPage 1 of 36Next