No technical background required
What happens behind the answer?
A simplified view of a language-model application. Click each stage to see what it does—and where it can fail.
Explore the system
Conceptual illustration, not a running model. Apps may also retrieve documents or use tools; training and generating an answer are different processes. Read IBM’s introduction ↗
A small glossary
AI, machine learning, and generative AI
AI is a broad field. Machine learning fits patterns from data. Generative AI creates outputs such as text or images.
Model, token, and context
A model is a learned computational system. Tokens are units used to process text. Context is the input available for a particular response.
Hallucination
A plausible output that is inaccurate or unsupported. Check claims against original sources. IBM explainer ↗
RAG and agents
Retrieval-augmented generation supplies documents as context. An agent uses a model alongside tools to take steps toward a goal. Neither guarantees correctness.
Open weights, open source, and local models
Open weights means model parameters are available. Open source also concerns licensing and other components. Local means running on your hardware; check network settings, licenses, and data handling.
Misconception: a high rank means it is reliable
A ranking reflects its evaluation method and participants. Test your actual task, factual accuracy, accessibility, and privacy needs. Explore Arena ↗