AI Glossary

An AI system's capacity to dynamically adjust its internal representations and behavioral strategies based on changes in context, task demands, and environmental conditions — without explicit retraining.

The degree to which an AI system's actual behavior conforms to its specified behavioral constraints — measured across a defined set of scenarios and expressed as a compliance rate.

The application of artificial intelligence to accelerate the identification, design, and optimization of pharmaceutical compounds — reducing the time and cost of bringing new drugs from hypothesis to clinical trial.

A processor designed specifically for executing trained neural network models — optimizing for throughput, latency, and energy efficiency at inference rather than training.

Adversarial testing of AI systems by teams attempting to find failure modes, safety violations, and harmful outputs — analogous to cybersecurity red-teaming but applied to model behavior.

The field studying how to build AI systems whose goals, values, and behaviors remain beneficial and consistent with human intentions as the systems become more capable.

Predicting the three-dimensional structure a protein adopts from its amino acid sequence alone — using computational methods to solve a problem that took decades of experimental effort per protein.

An AI system architecture that supports both fast, low-latency responses for routine queries and slower, more compute-intensive deliberative reasoning for complex tasks — allocating resources based on task demands.

AI that perceives and acts in the physical world through a body — integrating sensorimotor capabilities with cognitive intelligence to reason about and manipulate physical environments.

The use of mathematical proof techniques to establish that an AI system satisfies specified behavioral properties — providing guarantees rather than statistical estimates of safety.

An artificial general intelligence system operating at the current leading edge of capability — able to reason, plan, and act across diverse domains without task-specific engineering.

High Bandwidth Memory 3E — the current-generation stacked DRAM technology offering the highest available memory bandwidth for AI accelerators, achieved by mounting DRAM dies directly on the processor package.

The growing performance bottleneck caused by the gap between processor speed and memory bandwidth — particularly acute in AI inference, where moving model weights dominates computation time.

A hardware paradigm that places compute logic physically close to or within memory arrays — reducing the energy and latency cost of moving data between memory and processors.

Methods for protecting quantum information against decoherence and gate errors by encoding logical qubits redundantly across multiple physical qubits and detecting errors through syndrome measurements.

Neural network designs built from the ground up for quantum hardware — not classical models with quantum layers added, but architectures whose fundamental computational primitives exploit quantum mechanical properties.

A set of self-imposed commitments governing how an AI organization scales model capabilities — specifying evaluation thresholds, safety requirements, and deployment gates that must be passed before releasing more capable systems.

The process of training a robotic or physical AI system in simulation and deploying it on real hardware — addressing the performance gap between simulated and physical environments.

The hypothesis that a sufficiently complex computational system could exhibit functional properties associated with consciousness — including subjective experience, self-modeling, and integrated information processing.

A chip architecture in which computation is organized as a mesh of processing tiles with local communication — data flows spatially through the array, minimizing off-chip memory accesses for matrix operations.

A neural network architecture that maintains a persistent, updateable representation of spatial structure — enabling autonomous navigation and environment understanding without pre-built maps.

The use of pressure, force, shear, and vibration sensors embedded in robot hands and grippers to provide physical contact information that vision cannot supply — enabling dexterous manipulation and slip detection.

An energy efficiency metric for AI language model inference — measuring how many output tokens a system generates per joule of energy consumed. Higher is more efficient.

An organizational strategy in which a single entity controls the full stack of AI development — from hardware and infrastructure through models, training, and deployment — enabling tighter optimization across layers than component-by-component assembly allows.

Robotic manipulation of objects the system has never seen during training — generalizing motor skills to novel geometries, materials, and configurations without additional learning.