Learning German – Level B1.2: Explaining Large Language Models (LLMs)

Okay, so here I am in Berlin, trying to settle in and, you know, actually use my German beyond ordering a coffee or asking for directions. My TELC exam is looming – B1.2 Writing, specifically, which means I have to describe something. And they want me to explain how these Large Language Models work. Honestly, the idea seemed completely baffling at first. It’s not like I wanted to become a computer scientist! But after a few weeks of talking to people and actually trying to understand, it’s starting to make a little more sense – and thankfully, it’s relevant for my writing task.

First Steps: Conversations in the Café

The first thing I realised was that everyone talks about “Modelle” (models) all the time. Like, if you ask for a Kaffee (coffee), someone might say, “Das ist ein gutes Modell!” – “That’s a good model!” It doesn’t mean they’re praising your coffee-making skills! They are referring to these LLMs. I started hearing about them from my colleague, Thomas, at the Autowerk (car factory).

“Schau mal,” he said, pointing at his tablet, “ich habe das hier benutzt, um eine E-Mail an meine Frau zu schreiben. Es hat mir sogar Sätze vorgeschlagen! Es ist wie ein… ein Modell für Schreiben.” (Look, he said, pointing at his tablet, “I used this to write an email to my wife. It even suggested sentences! It’s like a… a model for writing.”)

This was my first clue – these models aren’t about completely creating things from nothing; they’re about assisting with writing. I started noticing that people would ask for help with their emails, letters to the Gemeinde (local council) – “Hast du einen guten Satz für ‘Ich möchte mich beschweren wegen…’?” (Do you have a good sentence for ‘I want to complain about…?’). The models provide options, suggestions.

Trying to Explain it – My First Attempts

So, I decided to try and explain it to my German teacher, Frau Schmidt. I started with something like: “Also, diese großen Modelle… sie haben viel Text gelesen, oder? Und dann können sie neue Sätze schreiben?” (So, these big models… they’ve read a lot of text, right? And then they can write new sentences?)

She smiled patiently and said, “Ja, richtig. Aber es ist nicht so einfach. Sie lernen aus Mustern. Stell dir vor, du lernst Deutsch, indem du viele Bücher liest und dann kannst du selbst schreiben. Das Modell macht dasselbe, nur in viel größerem Umfang.” (Yes, that’s right. But it’s not as simple. They learn from patterns. Imagine you learn German by reading lots of books and then you can write yourself. The model does the same, just on a much larger scale.)

That made a little more sense. It’s like learning to imitate – mimicking what I’ve seen. I realised I needed to be specific about how these models ‘learn’.

Vocabulary for My Writing – Schlüsselbegriffe (Key Terms)

Here are some phrases and words I’ve picked up that will definitely help me in my TELC task:

• Textdatenbank: (Text database) – They need huge amounts of data!
• Mustererkennung: (Pattern recognition) – They look for patterns in the text.
• Wahrscheinlichkeiten: (Probabilities) – They predict which words are likely to come next based on those patterns.
• Generieren: (To generate) – To create something new.

I even practiced saying these out loud while ordering Pizza (pizza). “Ich möchte bitte eine Pizza mit Salami und ich brauche Hilfe, um einen Satz zu generieren!” (I would like a pizza with salami and I need help generating a sentence!). The pizzeria guy just stared at me for a moment before chuckling and taking my order.

A Small Misunderstanding – The Real World

Last week, I was talking to my neighbour, Herr Müller, about the weather. He mentioned he’d been using an app to write his letters to the Zeitungsredaktion (newspaper editorial office) complaining about the noise. I asked him if it was a “Modell” and he looked at me completely confused.

“Ein Modell? Warum sollte ich ein Modell benutzen, um einen Brief zu schreiben?” (A model? Why would I use a model to write a letter?) He explained that he was just using an online tool – a very simple one! This showed me how easily the term “Modell” can be misunderstood. It’s not always about complicated AI; it’s often just about simpler tools that help with writing.

Focusing on Description for My Exam

Now, back to my TELC task. I need to describe how these LLMs work in a way that’s clear and understandable. I’m going to focus on the core ideas: they learn from massive amounts of text, identify patterns, and then use those patterns to predict what words should come next – generating new text based on what they’ve learned. I need to remember Thomas’s example with the email; that’s a good concrete illustration!

I think if I write something like: “Diese Large Language Models lernen aus riesigen Textmengen und nutzen Mustererkennung, um neue Sätze zu generieren,” (These Large Language Models learn from massive amounts of text and use pattern recognition to generate new sentences) – that will be good. Hopefully, it will get me a good mark! Ich drücke die Daumen! (I’m crossing my fingers!).

—