I answered your question before you asked it. You didn't like the answer so you arbitrarily claimed you were talking about morality rather than laws. Do you think setting booby traps is moral? If so, that's a reflection that you are not just a dipshit but a sociopath as well.

To be clear, you are a dipshit who won't answer basic questions to defend your awful analogy. If I warn you that being a dipshit might lead to getting shot, does that excuse anyone who actually shoots you?

The law in question here makes it a crime when someone "knowingly causes the transmission of a program, information, code, or command, and as a result of such conduct, intentionally causes damage without authorization, to a protected computer". Do you think that's satisfied by your sign hypothetical? I think you've left out some of the elements related to intent, and made the "causes the transmission" element much less clear. That's why I did not say anything like what you suggest.

Booby traps are still illegal and immoral even if you have "keep out - no trespassing" signs all over your property, and in this case it wasn't even the malicious abuser's own property.

Intentionally destroying somebody's data is still a crime if they have backups. The intent is the difference between this and equipment failure.

Do you think that women should go around in burqas and only have themselves to blame if they get raped when not wearing a burqa?

In this case, that's the person who snuck malicious instructions into the prompt, intending for it to destroy data.

This isn't rocket science. There are projects that do the same kind of thing conscientiously -- they have an AGENTS.md or a section in their README that basically tells an AI agent not to write code or create a bug report or whatever. Bad users can work around that, but those bad users are less harmful than the kind of person who prompt-injects malware.

The LLM companies are not causing the transmission of the command to delete things. If anyone, it's the jqwik developer who made test code generate the instruction at runtime in a hidden form.

The intent to damage is crystal clear here, and the attempt to hide the command makes it easy to assign responsibility for the knowing transmission of the command.

LLMs are unpredictable, but that's a very different threat than malicious instructions hidden by an attacker. The person who hid the malicious instructions is analogous to the hypothetical person who trained a dangerous dog and released it in a kindergarten.

Computers don't have agency; people do. The person who tells a computer to delete data is the one responsible for the deletion. This isn't rocket science.

My theory is that their employment of brain cells decreased by 20%.

Unemployment statistics count people before, while and after they collect unemployment benefits. The thing you called a trick is only your imagination.

https://www.bls.gov/cps/cps_ht...

Science doesn't "show us" what you claim. The article you linked is an editorial that is long on vague words like "some" and "many", but the only place it makes the claim you do is in a bit that is clearly presented as the author's opinion, with no evidence or study supporting the conclusion.

You're conflating concerns. Most government systems are required to log the hell out of their inputs and outputs. Making decisions to destroy something based on ephemeral data could happen just as easily on the ground as it could in orbit -- it has nothing to do with what kind of system (large neural network, traditional ML, human decision, or something else) makes the decision or where the decision happens.

I'm not sure what you mean. Do your eyes count as "long range sensors"? Mine can see stars that are many light-years away, and eyes are not a new invention.

They claim to have realized (invented) a better quantum magnetometer and a way to process the data to do a pretty amazing kind of detection. That's one very specific kind of long range sensor, with improvements over previous long range sensors but also limitations of its own. It's presumably not a magic device that Pareto dominates other long range sensors.

What's the differential latency of running a strong model for several turns (or the equivalent) on a spacecraft's power budget compared to a data center's power budget, especially once you factor in redundancy to manage single-event upsets in the huge RAM array needed for that model?

I use Claude rather than a local model because I don't want to wait all afternoon for the quality of results I can fit into 128 GB RAM.