Bavaria mulls creating its own version of ChatGPT

The argument is that the wildly popular large language model is based on American bias and technology

Last week, Bavaria’s Science Minister Markus Blume announced that he would seek the development of a Bavarian AI base model, which he tentatively called BayernGPT. He argued that this way the German state could stake its own technological independence, not linked to US-based technologies.

In his view, the current AI race is dominated by American tech giants, which invariably leads to training the large language models with American values, bias and knowledge. And Bavaria could forge its own path in this development.

BayernGPT

It remains a bit vague what a Bavarian AI model would look like for now and how its values will differ from other LLMs, but what’s clear is that the state government has an ambition to not fall behind on the developments affecting the world.

Last December, Bavaria’s Prime Minister Markus Söder announced that the authorities will fund the conversion of TU Nuremberg into an AI university, where a wide variety of applications of AI in life will be studied. This means that the BayernGPT project is just one pillar of an overall strategy for the state to move forward with the times.

The state government plans to use artificial intelligence “broadly” in administration. Its plan seems to be to declutter bureaucratic loads and to accelerate efficiency. To do this, however, the AI model has to be trained locally and the development of this task takes about a year at least. However, even before getting to this point a robust computer infrastructure needs to be set up and a roadmap developed. That process though could take around five years.

The finished basic model BayernGPT will then be published under an open-source license. This means that anyone can use and develop it without having to pay for it. This could be interesting for business and industry, which can adapt the basic model for their purposes.