The fundamentals and mechanics behind modern language models and AI workflows — from LLM basics through prompting, agents and RAG to local models, fine-tuning, evaluation and API usage.
AI Concepts is the topic field beneath the surface: not which model you pick, but how modern language models actually work and how you assemble them into reliable tools. Once you understand the mechanics, you make better decisions — and run into hallucinated answers, blown budgets and systems that shine in a demo but fail in everyday use far less often.
It starts with the LLM basics: tokens, the context window, next-token prediction and the difference between probability and truth. Prompting builds on that — steering the model through its input alone. Good prompts are not magic spells but clear instructions, examples and structural cues that guide the output into predictable paths.
A single prompt is rarely enough. Agents are autonomous, multi-step systems: the model plans, calls tools, reads their results and decides on the next step — until the task is done. So a model doesn’t only draw on its training knowledge, RAG (Retrieval-Augmented Generation) comes into play: your own documents are split into chunks, stored as embeddings in a vector database, retrieved to match the query and sorted via reranking before they enter the model as context. That way the AI answers based on your data, not on guesses.
If you need data protection or cost control, you look at local LLMs — here concepts like model formats, quantization (compression at the cost of minimal accuracy) and the right hardware matter. Training & fine-tuning goes a step further: a base model is specialised on your own data or tasks. And because quality can’t be guessed, evaluation belongs here — evals, benchmarks and LLM judges that measure whether changes are really an improvement. The technical frame is API usage: streaming for smooth output, caching to save on recurring contexts and rate limits you have to plan for in production.
Below you’ll find a topic world around AI concepts: current news on new methods and tools, blog articles with background and practice, lexicon articles for deeper insight and a glossary of the most important terms. Use the filters above to jump straight to a sub-topic — from prompting through agents and RAG to evaluation.
Anthropic discloses the numbers: Claude writes >80% of its production code. At the same time, the company argues for the option to pause frontier development.
Anthropic's first major threat report: the share of high-risk actors using AI nearly doubled. Plus one largely autonomous agent attack.
A US export-control order forces Anthropic to disable Fable 5 and Mythos 5 worldwide. In the EU the model stays unselectable — restoration unconfirmed.
Autonomous coding agents in an endless loop: what Ralph Loop and Loop Engineering really are — and why the bill explodes without hard stops.
2026 industry data: agentic AI burns 5–30x more tokens than a chat. What that means for multi-agent budgets.
At its one-year mark the A2A protocol counts 150+ organizations, ships in every major cloud, and reaches its first stable 1.0 spec for agent interop.
A typed keyword triggers a fixed AI routine — and every single step must be committed before the next one appears. Why that's the actual trick.
Headless agents on your own machine, fed by the subscription you already pay for instead of an API bill — how boostN orchestrates many models.
Three frontier models, the same 1000-line script, three different finding lists — and why that very spread makes multi-orchestration strong.
You write rules, curate memories, build guidelines — and a week later you're going in circles again. Why that is, and what actually helps.
Effort scales breadth, deep thinking scales depth. When each setting makes sense — with three clear examples and one rule of thumb.