Glossary  /  Data & development

LLM – large language models, explained

An LLM (large language model) is an AI model trained with billions of parameters to understand, process and generate language. It is built on the transformer architecture, which uses self-attention to recognise complex relationships in language and, on that basis, produces text or answers questions.

Data & development ·2 Min. Lesezeit ·Glossary

How does an LLM work?

Training & prediction

LLMs are trained on enormous volumes of text by learning to predict the next word (token) in a sequence. The result is a language model that can handle a remarkably wide range of tasks.

Transformer architecture

Through self-attention the model analyses the relationships between words in context. That is what allows it to process even long texts coherently.

Fine-tuning & RLHF

After pretraining, LLMs are often improved for a specific purpose through fine-tuning or reinforcement learning from human feedback (RLHF).

Fields of application

Chatbots & conversational AI

LLMs are the basis for modern chatbots that can hold human-like conversations.

Content creation

From summaries to blog articles, LLMs can generate a wide variety of content.

Translation & language services

LLMs translate text and serve as the foundation for multilingual features in applications.

Code generation

Specialised LLMs support developers in writing and reviewing source code.

Knowledge work & research

LLMs help analyse documents, literature and databases – frequently in combination with RAG (retrieval-augmented generation).

Benefits of LLMs

A generalist model

One LLM can solve many different tasks without being trained separately for each one.

Few-shot and zero-shot learning

LLMs can handle tasks successfully even with few examples, or none at all.

Adaptability

Fine-tuning lets you optimise models for industry-specific data.

Risks and challenges

Hallucinations

LLMs can output incorrect or invented content that nevertheless sounds convincing.

Bias & ethics

Biases in the training data carry over and can lead to discriminatory results.

Compute & resource demands

Training and inference require powerful GPUs and a lot of energy.

Data protection & security

Handling sensitive data in LLM environments has to be clearly governed and protected.

LLMs and centron

Running LLMs in production depends on strong, scalable and secure infrastructure. centron provides the foundation:

centron component Role for LLMs
Cloud GPU Training and inference of large language models on state-of-the-art GPUs
ccloud³ VM Flexible compute environments for model hosting and API delivery
Managed Firewall Secure access protection for LLM endpoints and data pipelines
Backup & Recovery Protection for training data, models and configuration
CI/CD Pipelines Automated rollout and updating of LLM workflows

FAQ on LLMs

What is an LLM?

A large language model is an AI model trained on billions of text samples in order to understand and generate language.

What are LLMs used for?

They are used in chatbots, text generation, translation, coding assistance and document analysis.

What are the benefits of LLMs?

LLMs are flexible, pick up new tasks quickly and can be adapted to industry-specific data.

What are the risks?

Hallucinations, bias, high resource demands and data protection and security questions.

Secure LLM deployment with centron

With ccloud³ VM, Managed Firewall and Backup & Recovery you build a robust foundation for your LLM projects – hosted in data centres certified to ISO 27001 on the basis of IT-Grundschutz.

Cloud GPU – start here ccloud³ VMs

Get started for free

Sign up and receive €200 credit at centron within your first 60 days.

This promotional offer applies to new accounts only. Available exclusively to businesses.