It seems to be fully prompt based, so the AI still can say anything it pleases.

How well do these complicated prompt systems usually work? My strategy is to stick mostly to just simple prompts with potentially some deterministic tools and vendor harnesses, based on the rationale that these are what the models are trained and evaluated with. And that LLMs still often get tripped up when their context is spammed with too much stuff.

This is at best a conjecture. It's arguable what terms are acceptable and to whom, but Russia has offered and is offering substantially milder terms than taking all of Ukraine. They probably wouldn't even want western Ukraine if offered.

International treaties tend to be always somewhat aspirational and are often violated and sparsely enforced. This does not mean they don't have any effect.

Personnel mines has been a topic of considerable public, political and military discussion about personnel mines in Finland for a long time and has been analyzed in depth. Even now, after a significant political mood change at around 2024, the official line (e.g. in the government proposal[1] that led to the withdrawal) is that personnel mines have mostly an auxiliary role in the defence strategy.

[1] https://www.finlex.fi/fi/hallituksen-esitykset/2025/56

But if that's the case, what are "all the force multipliers"? Chemical weapons? Biological weapons? What share of the GDP for defence?

Part of these kinds of treaties is to accept some additional difficulty or expenses in defence for a more widespread benefit. I'm living in Finland and I would have accepted these.

Discretizing e.g. time or space is perhaps a bigger issue, but the issues are usually well understood and mitigated by e.g. advanced numerical integration schemes, discrete-continuous formulations or just cranking up the discretization resolution.

Analytical tools for discrete formulations are usually a lot less developed and don't as easily admit closed-form solutions.

https://hannahritchie.substack.com/p/ai-footprint-august-202...

Author (in practice author institution, in practice with public funds) pays open access is less bad than locking articles behind paywalls, but it's still a racket.

This CERN system is about diamond open access, meaning that neither authors nor readers pay.

Do you have some specific techniques or strategies for LLM text detection? Have you validated them?

For me just disabling IPv6 has given the biggest payoff. Life is too short to waste time debugging obscure IPv6 problems that still routinely pop up after over 30 years of development.

Ever since OpenVPN silently routed IPv6 over clearnet I've just disabled it whenever I can.

Heroin (diamorphine) as a molecule isn't really substantially more problematic than e.g. oxycodone or fentanyl, which are in widespread medical use. The specific ban of heroin is more a historical contingency.

There seems to be a manufactured narrative from the US right how "Europe" is somehow doing large scale censorship.

It was XFree86 until around mid 00s after which the X.org fork took over.