In today’s highly competitive and data-intensive landscape, technical documents—especially patents—represent a strategic resource that is still largely underexploited. While recent advancements in Generative AI and LLMs have raised expectations across many sectors, their application in technical domains is still hindered by high implementation costs, unclear ROI, persistent hallucination issues and lack of control on the accuracy of the results. Studies report errors particularly in tasks requiring deep semantic and structural understanding—such as novelty evaluation or FTO analysis—where precision and recall are essential. LLMs, by design, tend to generate fluent but potentially misleading outputs, often failing to signal uncertainty when confronted with complex or unfamiliar content.

The talk will discuss the potential limitations of AI models and propose the combination of Artificial and Human intelligence as a possible strategy to enable reliable, scalable, and context-aware interpretation of complex technical content.