It might be a helpful abstraction to view LLMs as a compression/de-compression algorithm that can utilize an enormous storage of knowledge to make the process much more efficient, as long as you accept the fact that this a very lossy compression which only preserves the core concepts contained in the input but is free to change the representation of this information content. And, of course, it is prone to make wrong associations and hallucinate content.

How did our tool for “get situational awareness by asking the constituency questions” perform during the KSÖ exercise last week?

Not just the seasons, or my attempts to appear in the office in an outfit other than holey conference shirts, shorts and Birkenstock slippers that are cyclical. The desire of politicians for a "government trojan" or surveillance of digital communication seemingly follows a constant rhythm as well - and apparently it's that time again. Federal Chancellor Karl Nehammer is making the surveillance of digital communication a fixed condition for a future political coalition.

Back in January 2024, I was asked by the Belgian EU Presidency to moderate a panel during their high-level conference on cyber security in Brussels. The topic was the relationship between cyber security and law enforcement: how do CSIRTs and the police / public prosecutors cooperate, what works here and where are the fault lines in this collaboration. As the moderator, I wasn’t in the position to really present my own view on some of the issues, so I’m using this blogpost to document my thinking regarding the CSIRT/LE division of labour. From that starting point, this text kind of turned into a rant on what’s wrong with IT Security.