The NewsRupt

How models are trained, what they cost to run, and the vocabulary that keeps changing meaning.

• What is a large language model?A large language model is a neural network trained on very large amounts of text to predict the next token in a sequence. That single objective, applied at scale, produces a system that can summarise, translate, answer questions, write code and follow instructions. A model does not look answers up in a database at inference time; it generates them from learned statistical structure, which is why it can be fluent and wrong at the same time.
• What is an AI agent?An AI agent is a system in which a language model decides which actions to take, executes them through tools, observes the results and repeats until a goal is met or a limit is reached. The distinguishing feature is not autonomy in the abstract but a loop: the model''s output becomes an action, and the action''s result becomes the model''s next input. Reliability falls as the number of steps in that loop rises.
• What is context engineering?Context engineering is the discipline of deciding what information occupies a model''s context window on each call: instructions, retrieved documents, prior turns, tool definitions, examples and state. It replaces prompt engineering as the central craft once systems become multi-step, because the limiting factor stops being the wording of one instruction and becomes the selection, compression and ordering of everything the model sees.
• What is inference, and why does it cost what it costs?Inference is running a trained model to produce output. Unlike training, which is a one-off capital cost, inference is a recurring cost paid per request. Providers price it per million input and output tokens, and output tokens usually cost several times more than input tokens because each one requires a separate forward pass through the model. Reasoning models, which generate large numbers of hidden intermediate tokens, can cost far more per answer than their headline price suggests.
• What is retrieval-augmented generation?Retrieval-augmented generation is the practice of fetching relevant documents at request time and placing them in the model''s context so the answer is grounded in specific sources rather than in the model''s parameters alone. A retrieval step — keyword search, vector search, or both — selects passages; the model is then asked to answer using those passages and to cite them. RAG reduces, but does not eliminate, fabricated answers.

The layer underneath the applications: retrieval, serving, protocols and the trade-offs each imposes.

• What is a vector database?A vector database stores embeddings — numeric representations of text, images or other data — and finds the ones closest to a query vector. It exists because exact nearest-neighbour search over millions of high-dimensional vectors is too slow, so these systems use approximate indexes that trade a small amount of recall for a large amount of speed. They are search infrastructure, not a replacement for a relational database.
• What is an inference provider?An inference provider runs models on its own accelerators and sells access per token or per GPU-hour. The category spans three different businesses: serverless token APIs for open-weight models, dedicated endpoints with reserved capacity, and raw GPU rental. They compete on price per token, tokens per second, time to first token, model coverage and the contractual questions — data retention, region, uptime — that decide whether an enterprise can use them at all.
• What is MCP (the Model Context Protocol)?The Model Context Protocol is an open protocol for connecting language-model applications to external tools, data sources and prompts through a common interface. Instead of every application writing bespoke integrations for every system, a server exposes its capabilities once over MCP and any compatible client can use them. It standardises the plumbing between models and tools; it does not decide whether a model uses a tool well.

What happens when this technology meets procurement, compliance and an operating team.

• Why do enterprise AI pilots fail?Most enterprise AI pilots fail for organisational rather than model reasons. The common pattern: a pilot is scoped around a demo rather than a measured process, no baseline is recorded, ownership sits with an innovation team rather than the function that would run the system, data access is granted for the pilot but not for production, and no one has budgeted for evaluation, monitoring and human review. The model is rarely the binding constraint.

How answer engines retrieve, select and cite — and what can honestly be measured about it.

• How do AI systems choose which sources to cite?AI answer systems generally run a search, retrieve a set of candidate pages, select passages that appear to answer the question, and generate an answer citing the pages the passages came from. Selection therefore favours pages that are retrievable for the query, self-contained enough that a single passage answers the question, and corroborated elsewhere. Citation is a retrieval outcome, not a ranking reward, and it varies between runs of the same question.