Definition
Large Language Models (LLMs) are neural models — typically transformer-based — trained on large text corpora to predict and generate language. They power chat assistants, coding agents, retrieval pipelines, and domain-specialized tools.
Key Points
-
2026-07-30: ICML paper argues role-confusion is structural — models trust style over role tags; cot-forgery raises jailbreaks to ~60% (2026-07-30-cot-forgery-arxiv)
-
Built on transformer-architecture and related foundation-models / language-models
-
Domain and language-specialized variants (e.g. kotlinllm for Kotlin) improve IDE and agent coding quality
-
Tool use and protocols such as model-context-protocol connect LLMs to external systems
-
Safety, evaluation, and governance remain central product and policy concerns (ai-safety, ai-governance)