Large Language Models: A guide through the jungle of AI technologies

With the release of ChatGPT and the growing interest in Large Language Models (LLMs), a fast-moving market has developed whose dynamics and diversity can be challenging. In this article, we would like to compare the most important models and help you to keep an overview in order to select the right model for your company.

Large Language Models: More than just ChatGPT

LLMs are not a reinvention of OpenAI. Back in 2017, Google and the University of Toronto laid the foundations for today’s generative AI with their Transformers architecture. This technology makes it possible to generate complex texts and dialogs by understanding relationships between words over long text sequences.

OpenAI set the first milestone in 2018 with GPT-1 and has driven development up to GPT-4, with an estimated 170 trillion parameters.

Licenses and rights of use: a complex field

Choosing the right model depends heavily on transparency and usage rights. Many models based on open source are not suitable for commercial purposes. For example, some use data from ChatGPT, which can cause legal complications. Other models, such as Falcon and LLaMA-2, have adapted their licenses to enable broader use.

Data security: a decisive factor

Data security plays an important role in the selection of LLMs. German start-ups such as Aleph Alpha focus on data protection and operate their own data centers in Germany. OpenAI and Microsoft have also improved their privacy policies to give users more control over their data.

Map of the Large Language Models

There are numerous models on the market. Your choice should be based on your specific requirements:

Data-sensitive projects: A self-hosted model is ideal here.

Less sensitive data: OpenAI or Azure OpenAI are good options. The latter offers additional compliance benefits through regional hosting options.

Long contexts: Claude offers itself, with better API availability and lower latency compared to GPT-4.

Ethically correct answers: LLaMA-2 could fit, thanks to measures such as Supervised Safety Fine Tuning and the use of a Safety Reward Model.

Conclusion

Choosing the right LLM for your company depends on many factors, from data protection and licensing regulations to the specific requirements of your use case. With the right knowledge, you can make an informed decision that will help your company to make the most of the potential of AI.

Let’s start the dialog: What experiences have YOU had with LLMs? Share your thoughts and experiences with us!