In the Stuxnet attack, famously [0], the malware propagated via craftily-formed autorun.inf on USB sticks that ferried data into the facility. Nothing crossed the otherwise-robust air gap in the other direction, but nothing had to in that case.

Lessons have been learned since then, but this sort of possibility (electromagnetic exfiltration vector) has, for quite a long time, been taken seriously enough to inform secure facilities’ physical design [1].

[0] https://www.eweek.com/security/stuxnet-worm-nine-facts-every...

[1] e.g. https://www.nsa.gov/portals/75/documents/news-features/decla... , https://www.ncsc.gov.uk/information/tempest-and-electromagne...

Maybe we can use AI agents to patrol the boundary for the AI agents. After all, the only thing that stops a bad guy with an AI swarm is a good guy with an AI swarm…

Ah! Of course:

> Teams of frontier-model agents were tasked with finding vulnerabilities and misconfigurations that could allow them to take unauthorized actions […] Accordingly, we’ve made high-compute red-teaming part of the safety cases we use to evaluate whether our frontier training and evaluation runs can proceed securely.

I think you nailed it with respect to the fading magic of scientific culture, and the power of shared purpose to gloss over social differences.

Even as a plot device, it feels like the notion that {solving material/energetic needs solves existential needs} feels of a piece with a very particular, commoditized, late-industrial type of social climate. That “tea, Earl Grey, hot” would mean the same thing every time, down to the molecule—to an appreciator of tea, no less—always struck me as a dismal proposition.

Everywhere I look, though, I feel like, as soon as one source of friction gets greased, people forge right ahead and find (or create) a new site of friction.

Food security’s largely solved? What a challenge dieting can be! Not to mention toxins and gluten and agricultural monoculture and the livestock’s moral standing!

Magic robot writes all my code for me? Filthy BigAI, limiting my tokenmaxxing strategy to only billions of words of thought per week! And they want me to pay them an amount approaching the cost of a shared desk in an office?! Plus, the robot sometimes makes mistakes that require me to eyeroll _read_ the code…

Or, more benignly: I think of my skater friends. No sooner do they nail one demanding or stupid trick, than they’re reshuffling the obstacles and pushing their flips an extra revolution. A 4-minute mile was revolutionary, now it’s routine. Everest became packaged, so the wealthy seek out lesser but unconquered peaks [1].

No, we’ll have no problem finding and inventing friction and forming ad hoc communities accordingly. To the extent the friction is collectively-initiated rather than imposed by Nature red in tooth and claw, I’d say that’s probably a blessing—just as the unifying friction of the scientific frontier is more wholesome than the unifying friction of war or siege or famine.

> But a convincing replica of art might be all we need to sabotage the real deal, either through spam or demotivating the novice artists.

I share your worry here. I try to reassure myself on both points. On spam, I tell myself that breadth-of-attention isn’t the same as quality-of-attention: I’d imagine orders of magnitude more humans know Justin Bieber’s work than have ever known Mozart’s 6th. And as far as demotivation: I tell myself the frontier has always been a lonely place to operate. Great artists famously die poor and unknown: their urge, their obsession to create, seems to be almost possessing; to irrational degrees. Most people follow, a few lead; the greats in the past did so at even higher cost than those today.

Thank you for your thoughtful writing, here and around HN. Browsing through your comments you seem completely lovely and I’d love to invite you for a meal if we ever find ourselves in the same part of the world :)

[0] in which our hero bemoans his flagging stock of Whuffie. In Kissimmee. https://craphound.com/down/Cory_Doctorow_-_Down_and_Out_in_t...

> in the old days, if you were broke but respected, you wouldn’t starve; contrariwise, if you were rich and hated, no sum could buy you security and peace. By measuring the thing that money really represented—your personal capital with your friends and neighbors—you more accurately gauged your success.

[1] e.g. https://archive.is/9nuSv (Bloomberg)

I don’t really know or care how accurate or rigorous the analysis is (then again I’m obviously unskilled in the art of technical trading). What I’m missing is: does anybody really care about what’s being traced here?

As a relative outsider, I assume raw stock prices to be largely indicative. That is, their informative value is relative to that particular stock over time. And thus I imagine the $1-warning thing has more to do with the exchange specifying a minimum increment, a range of resolution in which their bids and offers allow enough nuance for the market forces to do their work. If the thing’s priced at 2 cents and I can only bid 1 cent or 3 cents, that’s not super useful or informative since it involves a 50% change in value on either side.

The new and specific facet of my AI frustration here, is that, true or false or accurate or hallucinated, the analysis leaves me none the wiser about whether I’m totally off base with those priors.

My specific discomfort feels a little different from asking the article to do something beyond what it’s setting out to do—it’s more that it feels like a psychic’s “cold read.” It could exist in a world that works the way I think it does, or in a world that works any number of different ways, and it would still feel like a plausible thing for somebody to have undertaken.

One place to look might be ministry or service. Illness, loss, grief, infirmity, death, age, family strains—people require care and people give care regardless of their postgraduate endeavors. Nuns and Jesuits and the like aren’t for everybody—but look to the neediest among us and, in large parts of the world, those sorts of order are the people doing the invisible, unglamorous, uneconomical, “chop wood, carry water” form of devotion and social labor day in and day out. In intimate community even with very difficult (and sometimes ungrateful) members.

Another might be the sort of ad hoc communities that arise in exigency: disaster, forced migration, encampments, solidarity against forces of difficulty.

Unhoused, unemployed, and retired people can provide a particularly illuminating contrast to the pastimes and concerns that animate a working life. A visible minority fill their days with addiction, but a whole lot are complete and interesting people with the full range of human ambitions and desires and fears and joys, and somehow they fill their waking hours with more than just material striving.

Another might be organizations like union halls, advocacy groups, music scenes, circus folk, jugglers, running clubs, political groups…

Kids are great for this too. They do a lot of hanging out, with nothing more than imagination and a drive to self-expression.

Even our hominid ancestors do a pretty good job of hanging out litigating social hierarchy and nurturing one another—no university degree or novel engineering task required…

If the claim is that some of the material there evidences police abuses of some sort, you can’t really let the offender pick the stuff they feel supports their claim and wipe the stuff that might weaken it.

And if there’s criminal material mixed in with the other, non-offending data that the offender claims they need, it seems like you kind of need some kind of adjudication process on an item-by-item basis.

They’ve clearly worked out a way to manage custody, access, and review for the purposes of the initial prosecution. Let the offender fund it under the same terms until they either fish or cut bait. Maybe if their claim of police abuse works out, this might be among the costs the court could award.

https://archive.is/xjSZA

Though it seems Yahoo was the more aggressive offender

The first spatter canvas was a new conceptual move and valued more for that than for the paint and canvas; subsequent works tended to ring hollow and command neither interest nor value. The first “ChatGPT, reword this like a pirate shanty” was worth a laugh, now it’s worth an eyeroll.

For that matter, as artists and writers incorporate these new tools into their practice, there’s still a difference between low-effort slop and the products of artistic obsession. AI image gen is commoditized right into the ground, yet you can still find interesting and distinctive works in, say, Midjourney’s community galleries. Or in at least one example from TFA:

> An AI reading machine uses computer vision, OCR, and language processing to find short poetic combinations in physical books, erase the surrounding words, pair them with images, and publish new volumes.

That sure sounds like an artistically interesting endeavor to me, independently of their choice to execute their concept through AI models.

https://.mozilla.org/showbug.cgi?id=2083449

Comments URL: https://news.ycombinator.com/item?id=50175062

Points: 20

# Comments: 0

I’m not sure how exactly that generalizes to LLM behavior, but to you as a human cultivating discernment, I imagine it implies that anything Internet-mediated (and unvetted by you) is less reliable than it used to be.

I understood it to be more of a finishing feed than a long-term dietary commitment on the part of the snail farmer, but both places I attended have since gone out of business. So maybe the economics are in fact against it…

More to the point, there’s a parallel system that came out of the same lawsuit that found the subway system illegally inaccessible to people with mobility needs. Those people can use MTA-funded “Access-a-ride” service, which offers private, scheduled door-to-door transportation for the price of a subway ride. It’s a really expensive way to do it, but a remarkable lifeline for the people I know who, for example, need occasional transportation to and from medical appointments.