Hacker News new | ask | show | jobs
by gyulai 118 days ago
> The EU trails the US not only in the absolute number of AI-related patents but also in AI specialisation – the share of AI patents relative to total patents.

E.U. patent law takes a very different attitude towards software patents than the U.S. Even if that wasn't the case: “Specialisation” means that no innovation unrelated to AI gets mind share, investment, patent applications. And that's somehow a good thing? Not something you can just throw out there as a presupposition without explaining your reasoning.
3 comments
dash2 118 days ago
> “Specialisation” means that no innovation unrelated to AI gets mind share, investment, patent applications. And that's somehow a good thing?

I don’t think the authors claim we should have 100% specialisation. They just say that the fact that the EU has fewer AI-related patents as a proportion of the total (less specialisation) is evidence that it is behind in AI. That seems reasonable.

link
direwolf20 118 days ago
Or it's ahead in not–AI.
link
eternauta3k 118 days ago
EU firms don't file EU patents necessarily, but rather in the relevant countries (including USA).
link
amelius 118 days ago
Makes me wonder how AI will influence the work of patent officers.

Perhaps it will make patent trolling a bit harder because it is easier to look up existing work and to check if an idea is obvious?

link
embedding-shape 118 days ago
> Perhaps it will make patent trolling a bit harder because it is easier to look up existing work and to check if an idea is obvious?

Haha, funny :)

No, it'll be like the rest of the industries that use more AI, they'll spend the same amount of effort (as little as possible) and won't validate anything, and provide worse service, not better. AIslop is everywhere, and seemingly unavoidable for companies to use more and more to cut more corners.

link
cor_NEEL_ius 118 days ago
The validation point is real. We tested this with AI presentation tools specifically - gave 6 of them the same prompt and fact-checked every claim against primary sources. Best accuracy was 44%. Most were under 20%.

The pattern was consistent: the tools produce confident, well-formatted output that looks thoroughly researched. But more than half the statistics were either distorted or completely fabricated. The worst part was finding the same fake stats appearing across multiple tools - not because they independently verified anything, but because they all absorbed the same bad data from training.

The productivity gains from AI are real, but so is the validation cost. People just aren't accounting for it.

link