I am not the least bit confused. The same technique that might be used to remove "harmful" behavior can be used to alter any behavior. That's self-evident to anyone with even passing concept of how models work.

False. The depth depends on the effort made to analyze the model and test behavior. There is no limit here: abliteration can be as shallow or as deep as one has time and money to spend.

Nope. It's a border thing: when you enter, you submit. The US Constitution is not a suicide pact, and it does not apply to every soul on the planet at all times.

Best you understand all that before you choose to tote your incriminating digital material through a border checkpoint, and/or destroy whatever it is you don't want seen.

Thems the rules, sport.

Inclusive of this case. This is outrage bait. It means nothing, and it will change nothing.

"Someone" already has, and there is a word for it: Abliteration. Manually hacking weights to bias behavior and create triggers for specific purposes, and there is no hidden prompt text or other demonstrative evidence this has been done. The only way to detect this is by comparing the abliterated model to the original.

And our children will live, Mr. Beale, to see that... perfect world, in which there's no war or famine, oppression or brutality. One vast and ecumenical holding company, for whom all men will work to serve a common profit, in which all men will hold a share of stock. All necessities provided, all anxieties tranquilized, all boredom amused. [emphasis mine]

-- Arthur Jensen, Network (destroying his golden goose)

Further viewing.

Pretty good question. Thing to do now is: see who complains and start watching them.

Elice (South Korea) shares plunged 30% on IPO. LOL.
Firmus (Australia) abandoned its planned $5 billion IPO
AirTrunk (Singapore) postponed its REIT
Microsoft threw Nscale (UK) in the trash.

This is all this morning. The Microsoft news appeared yesterday.

It many not be necessary, but certainly the instruments available on Earth to analyze samples are vastly more capable than whatever the lander has, given that it's only 5 tons and 90% of that mass is power/navigation machinery and propellant.

Production and shipping can be halted and the component inventory can be seized as evidence and the staff are subject to US law. All the FCC or DOD or whomever needs is a court order and some cops to drive to the site; no need to involve DHS or the State Department or shipping companies or longshoreman unions or any other bullshit. No need worry about all the angry senators and their foreign masters when tearing apart a dozen cargo containers on the docks and causing collateral damage to unrelated parties.

It's just a lot easier problem to deal with, even if all that actually happens is, as you say, assembly and boxing.

Why? This is what security people are supposed to be doing.

File this under: Do Your Job Please.

Brings to mind dieselgate, Madoff, Boeing, gasoline powered alarm clocks and all of the many, many other cases where the vast horde of lawyers we employ for government positions are basically just pencil whipping stuff and taking zero proactive steps to ensure any sort of actual security or compliance.

If you've ever replaced large surface mount ICs on a PCB, you'll know you don't want to rely on the result for anything that matters, or take the support calls and complaints for the ones that don't work quite right afterwards.

The Raspberry PI wallet vote is dominated by customers that have no reason to toy with replacing RAM chips. They want consistent quality from reputable sellers, and they appreciate that a door has been closed to Asian board rework houses, filling the supply chain with unqualified parts.

Honestly, that's what I want from Raspberry PI as well. I know the RAM is soldered when I buy it. I have nothing to complain about. When I want something less carefully controlled, cheaper and more flexible, there are ample choices there as well. Raspberry PI need not compromise their standards for that purpose.

The last few years, my Raspberry PI use case is a high-end ADS-B feeder: mast-head pre-amp and filtering, low loss coax, high gain antenna, 14-bit SDR, properly powered and grounded, running in non-room-temperature conditions, 24x7x365. The software is qualified, running on qualified hardware, and it all works flawlessly: every update works perfect, every power-outage is recovered from, etc. That's what I want from Raspberry PI and I'm glad they are there, providing a high quality, reliable platform that others can target without troubles and issues.

Excessive: beyond necessary, proper and/or normal.
Peace: an interval of time between wars, such as you've likely indulged as normal your entire life, whether you know it or not.

"Suffer" is equally important, for what should be obvious reasons. In peace, vast quantities of bullshit accumulate. The GP considered the benefits that some bolide might impart by setting it alight. I merely relate that a highly effective alternative that we once benefited from has been suppressed.

Indeed. LLMs are like a cryptocurrency tumbler for intellectual property. It is a metaphysical certitude that, sooner or later, a major legal challenge is going to happen with Linux and/or other open source software collections from some rent seeking copyright holder(s) due to AI contributions. Linus is taking a real risk here.

May you live in interesting times. Whatever outcome emerges from the forthcoming years long legal processes will set profound precedents.

I predict this won't actually start with Linux, however. It will be Adobe. There is a cottage industry of Photoshop clones being created with LLMs. Adobe is going to take someone to court over this.

Nah. No need to reach for anything that exotic. A couple good conventional wars would fix almost all of this. The best rationale I've discovered for a global ban on nuclear weapons is providing survivable opportunities to limit the excessive peace we currently suffer.