A growing AI vocabulary is making terms such as AGI, AI agents, chain of thought and inference increasingly common outside research labs. An updated AI glossary from TechCrunch now adds newer concepts including opaque recurrence, recurrent depth, neuralese, Model Context Protocol and RAMageddon alongside established terminology.
Many of the definitions describe how modern AI models are built and operated. Large language models use neural networks with billions of numerical parameters, or weights, while training teaches those models patterns and inference refers to running a trained model to produce predictions or responses.
Techniques such as fine-tuning and transfer learning adapt existing models for more specific tasks. Distillation uses the outputs of a larger teacher model to train a smaller student model, while reinforcement learning trains systems using feedback or rewards for successful behavior.
Agents and Infrastructure Get New Vocabulary
AI agents differ from basic chatbots because they can perform multistep actions, such as operating software, managing code or interacting with third-party services. Coding agents apply the same concept specifically to software development by writing, testing and debugging code with reduced human involvement.
Model Context Protocol, introduced by Anthropic in 2024, provides a standard way for AI systems to connect with external tools and data. API endpoints provide another mechanism for software and AI agents to communicate directly with outside services.
Infrastructure terms are also becoming more prominent. Compute describes the processing resources behind AI, while token throughput measures how much model workload can be processed within a given period.
Mixture of Experts models divide a network into specialized components and activate only selected parts for each request. Memory caching, including KV caching, reduces repeated calculations during inference to improve speed and efficiency.
Astra Introduces New Reasoning Terminology
Recent additions reflect changes in advanced reasoning systems. OpenAI introduced GPT-6 Astra on September 3 through an official release, bringing renewed attention to a technique described as opaque recurrence.
Opaque recurrence repeatedly passes information through internal model layers rather than relying entirely on readable step-by-step reasoning. Recurrent depth is the more technical name for the same underlying technique, while neuralese describes a hypothetical system that reasons entirely through internal numerical representations rather than human-readable language.
Other terms cover persistent limitations and industry pressures. Hallucination refers to AI generating incorrect information, validation loss measures model performance during training, and RAMageddon describes rising memory-chip shortages and prices linked partly to demand from AI infrastructure.
The glossary also covers long-running concepts including AGI, deep learning, diffusion, GANs, open source models, recursive self-improvement and parallelization. Definitions remain contested in some cases, particularly AGI, which OpenAI’s charter defines around highly autonomous systems outperforming humans at most economically valuable work.
Featured image credits: Magnific.com
For more stories like it, click the +Follow button at the top of this page to follow us.