Why is Google's AI so fucking funny? I actually ran out of usage just asking it the dumbest questions I could think of. I didn't even know that could happen.
Apparently all of these “AI is escaping containment” stories are actually just AI researchers not knowing jack squat about the absolutely most basic security practices.
Here's why I like this response (I got one more along these lines but not this detailed).
My whole criticism of METR is that that they're a collection of LW forum posters whose goal is to stop AI development, & to do this they've stood up a benchmarking & safety org that A) aims to make the already decided (to them) case that AI is moving too fast & is unsafe, & B) is not staffed by people who know anything about benchmarks or safety as an actual discipline.
This post essentially concedes the two-part (A & B) motivation criticism, i.e. that the whole plan here is the Pause, & they don't need to engage w/ legacy tech safety concerns because the plan is to Pause & then learn about that stuff & apply it to AI.
The post also sort of implicitly supports my first criticism (that they're a bunch of LW posters who cooked up a plan to Pause AI) by responding to a criticism of METR with essentially, "hey there! LW posters DO talk about this stuff!"
At any rate, the core disconnect between me & this person I'm QTing & METR & most of LW, is that I am in the "Normal Technology" camp. Also, normal tech can be extremely disruptive & dangerous, which is why safety matters, & the approach I'm criticizing (& that he's describing!) is inimical to safety IF we're dealing with a normal tech.
But if LW is right and AI really a totally not-at-all-normal tech & we're 10yrs away from Dyson Spheres (per Cotra in other threads), then ok yeah I suppose any and all types of resistance are licensed, including standing up a benchmarking & safety org that has very little connection to anything anyone else can recognize in the pre-existing areas of benchmarking and safety. So in that respect, I get it, tho I strongly disagree.
At any rate, it really does matter where you fall on this matter of "This is an emergency!" or "This is a dangerous but normal tech!", for exactly the reasons I'm talking about here. Because if one of those is true, then the other one is unsafe.
Show more
happy petrov day to all who celebrate
a few days ago I asked Opus 5.5 to write me a CSRF checker
it gave me one that failed open if it didn't receive a cookie
if you pause for a moment to try and read the code Astra (and I presume fable) are writing it becomes clear that “corrigibility” has become a matter of faith. they are using crazy meta-programming and abstruse primitives to write hyperefficient code
Show more
hot take: in practice, A/B testing is mostly unethical
this works especially well if you pronounce "Polycules" like "Hercules"
Polycules attack you like this when you’re a single person who just moved to Portland
5 DAYS LEFT
“It’s an excellent novel and exactly what the Prometheus folk have always said they want to reward.” — John Barnes, author of Directive 51.
Show more
I am a foaming political extremist so if your Political Quiz makes me answer 'neutral or it depends' to ten questions in a row it probably isn't a very good classifier
okay, I also hate the age verification thing but could you please stop tweeting cartoon gore pictures of the discord mascot into my timeline
i never considered that you used to have to send a letter to the newspaper to make this sort of tweet
This tweet sparked the beginnings of a thought in that modernity is an incredibly subjective and selfish era while wearing the mask of objectivity.
Of course all eras were selfish to some extent but it seems that in the past, people were aware of this. Chivalry and codes of warfare were produced amongst European duchies and dukedoms who went to war for political reasons. They pursued selfish aims, knew they were selfish and political, and thus were able to turn it into a bounded game. The bloodiest wars were those about “good and evil” or “right and wrong” but war is never really fought for this reason. Modernity treats all warfare as friends of humanity versus enemies of humanity.
Marriage, among the aristocracy or merchant classes, was settled in terms of titles and dowries. Love was not absent but at least they recognized there was some practical aim in mind. All these considerations are still there, but rather than being honest about them, with others or even with ourselves, we pretend it is a question of love or morality rather than our own interests.
All these practical, selfish motives still exist, yet we refuse to acknowledge them. We can say that the modern man is less integrated with his shadow in Jungian terms. Refusing to acknowledge the shadow does not make it go away, instead it makes your own behavior incomprehensible to yourself. We pursue the same ends as past eras but refuse to acknowledge to ourselves that we have some selfish interest be it politically or tribally or personally. Everything must be dictated and stated in objective terms of absolute morality, but language does not change reality, it merely obscures it.
The problem, of course, is that our own minds and own motivations are a large blindspot for ourselves — we often take our cues for these things from society at large, and it takes a great deal of introspection and interoception to parse truth from fiction. Thus, a world that refuses to acknowledge its shadow, or its own aims, and instead preaches this sort of fake objective dignity or whatever, dooms the overwhelming majority of its inhabitants to misery because they are not seeing reality for what it is, and ruining their own lives for the illusion they’ve been sold and have come to believe.
Show more
Many people read my tweets and assume I’m against all immigration, or against immigrants as a whole. That is categorically false.
The Japanese may come and go as they please.
one obvious takeaway here is that we should be more cautious about lying to AIs for training purposes