GPT-NL's copyright restrictions undermine Dutch sovereign AI ambitions

Summary

The article critiques the GPT-NL project, a Dutch sovereign AI initiative, for its overly restrictive approach to copyright compliance. By refusing to train on copyrighted data even where copyright law permits it, GPT-NL produces a model that cannot compete with frontier AI models. The piece argues that this approach misunderstands the information ecosystem — the real issue is not the right to refuse use of content, but the right to be paid for it. The abrupt removal of Anthropic's Fable model in the Netherlands triggered public debate about dependency on US AI models, with some commentators unfairly bashing GPT-NL's performance. However, the author contends the real failure is structural: GPT-NL's self-imposed copyright constraints cripple its capabilities from the start.

Source

bskyGPT-NL's copyright restrictions undermine Dutch sovereign AI ambitionsopenfuture.eu