Was ein Modell groß macht

A large language model is a transformer trained on huge amounts of text, with billions of parameters that store patterns of language. 🤖

The model's parameters are the learned numbers that shape its predictions. More parameters give it more room to capture subtle patterns.

During pre-training, the model reads vast text and learns by predicting the next word over and over, billions of times.

That simple goal is powerful: to guess the next word well, the model must absorb grammar, facts, and reasoning hidden in the text.

Scaling up means more parameters, more training data, and more compute. Growing all three together is what unlocks new ability.

Past a certain size, models show emergent abilities like translation or reasoning that smaller versions simply could not do.

A single large model is a general tool: the same weights can summarize, answer questions, write code, and translate.

These broad pre-trained models are called foundation models because you build specific applications on top of them.

After pre-training, models are fine-tuned on instructions and human feedback so they follow requests and stay helpful.

An LLM can only read so much at once. That limit is the context window, measured in tokens, and it caps how much you can send.

Even huge models can confidently invent facts. This is called hallucination, and it is why you verify important outputs.

What core task is a large language model trained on during pre-training?

Large models scale parameters, data, and compute. That scale brings emergent skills, but also a context limit and the risk of hallucination. ✅