AI Fundamentals

ChatGPT, Claude, Gemini, Joule — they are all powered by large language models. But what is an LLM actually doing when it responds to you? This post explains what an LLM is, how it is trained, what tokens and embeddings mean, and why LLMs predict rather than think — clearly, without the academic language.

AI hallucinations are not a bug that will be fixed in the next model update. They are a structural consequence of how large language models work. This post explains what hallucination actually is, why it happens at a technical level, the different types, and what practical steps reduce it — without the hype or the panic.

RAG is the most important technique for making AI reliable in enterprise settings. It gives an LLM access to your documents, data and knowledge at query time — so it answers from evidence, not from memory. This post explains what RAG is, how it works step by step, where vector databases fit in, and how SAP uses it in Joule and AI Core.

Chatbot, copilot, agent — these three words get used interchangeably in 2026 but they mean very different things. This post explains what an AI agent actually is, what separates it from a simple chatbot, how the planning loop works, what tools and memory do, and where agents succeed and fail in practice — without the hype.

Anthropic released MCP in November 2024. By March 2026 it had 97 million monthly SDK downloads and support from every major AI vendor. This post explains what MCP is, the problem it solves, how it works architecturally, and why it matters for anyone building or integrating AI systems — including in the SAP world.