Don't worry, I would be immensely impressed if they even finish a pretraining run for a competitive model. Let alone get to the stage of doing any kind of fine tuning for any kind of purpose.

Well, perhaps they could take Deepseek, feed it with all EU directives, ask on each if it's related to AI (or whatever) and spill out results.

It could be even nice idea for startup. All data are publicly available...

Here's a summary of what one of the members has done before.

https://www.ai.se/en/ai-labs/natural-language-understanding/...

They've cooperated with a research agency known in part for their Prolog implementation, i.e. they've been at it since the last massive "AI" hype cycle.