Was sind große Sprachmodelle?

Large Language Models (LLMs) power tools like ChatGPT. They are AI systems built to understand, generate, and process human language at the frontier of today’s AI.

The "large" in LLM means two things: billions or trillions of learned parameters, trained on enormous text datasets. That scale is what captures complex language patterns.

At heart, an LLM is a statistical model that predicts the next token in a sequence — like a wildly advanced autocomplete building sentences word by word.

The headline skill of LLMs is generating human-like text: stories, emails, summaries, code snippets, and natural conversation — making them powerful for creative and practical work.

Beyond writing, LLMs summarize long text, translate between languages, answer questions, and complete passages — versatile assistants for nearly any language task.

Key reality check: LLMs do not truly "understand". They are statistical engines that reproduce patterns from training data — no consciousness, emotions, or beliefs.

A major limit is hallucination: the model confidently states things that sound plausible but are flat wrong, inventing details when it lacks real information.

Another limit is the knowledge cutoff. An LLM only knows data up to its last training date — ask about recent events and you get stale or wrong answers.

Because LLMs learn from human text, they absorb its biases too — so outputs can reflect societal bias. Mitigating it is an active area of research.

Which of the following is a core limitation of Large Language Models (LLMs)?

LLMs are powerful but imperfect: brilliant at language, yet just statistical models prone to hallucinations, knowledge cutoffs, and bias. Knowing the limits is how you use them well.