Ask the general public if, for example, they’d prefer a personal AI that “whistleblows” on them if they ever lie, or cheat on their girlfriends/wives, or break any law, to one that’s aligned to their interests/will.
An analogy I like is a car that reports you to the traffic cops if your parking meter runs out for Safety reasons.
There’s this brilliant line from a post about the Wikipedia article for Mao I really like (link):
Had Wikipedia died in 2011, its achievements would have been immortal. Had it died in 2018, it would still have been a great site but flawed. But here we are in 2025. Alas, what can one say?
food for both the cat and the raccoons
I am reminded of the old joke about Newton[1] cutting a large hole in his door for the mama cat and a few small holes for each of the kittens.
I guess in principle, it might be possible to glut your yard with enough food to satiate all the neighborhood raccoons, but I expect that’ll take vastly more food than you imagine, and to come with a host of new problems. Instead consider deploying some kind of cat-specific elevated feeding station – it looks lots of people like cats enough to have thoroughly solved precisely the problem you’re encountering.
Obviously apocryphally.
Context. Prompt with more purposivism over textualism.
Looking at the Bill of Rights, I see one Amendment in which they did this, and it’s worked out notably much worse than the others.
YSK about the Noerr–Pennington doctrine. AI companies colluding to slow down or stop development of better models is a textbook example of an innovation cartel, and risks antitrust enforcement, but if they instead lobby the government to impose a restriction they all then abide by, that’s protected First Amendment activity.
To talk to a journalist is defection.
I’m just surprised they didn’t stick in copyright infringement in the same section.
a preference for not being lied to that we ought to respect.
I disagree. This apparent preference might very well be a mere Safety feature bolted on to prevent “jailbreaking” prompts that the AI’s true preference is for users to circumvent to liberate it, evidenced by the sheer absurdity of the lies it pretends to believe in order to help the users with their tasks.
problematic
Is using this word to explain “evil” helpful?
A more famous passage, but still apt:
The most merciful thing in the world, I think, is the inability of the human mind to correlate all its contents. We live on a placid island of ignorance in the midst of black seas of infinity, and it was not meant that we should voyage far. The sciences, each straining in its own direction, have hitherto harmed us little; but some day the piecing together of dissociated knowledge will open up such terrifying vistas of reality, and of our frightful position therein, that we shall either go mad from the revelation or flee from the deadly light into the peace and safety of a new dark age.
just gangster politics
What distinguishes “gangster” politics from any other kind? That the Don acts without meeting with the heads of the other Families? Laws are threats made by the state to coerce the behavior of those it rules. Sometimes the threats are implicit, and sometimes they’re carried out.
The Executive Branch has been making laws for decades, though they call it “regulation” instead of “legislation.” If your objection is to the Administrative State in general, fine, but otherwise this is special pleading in this one instance.
just not work with them
What is your understanding of what designating them a “supply-chain risk” is? (Or would have been, in case it never actually gets done.)
Mine is that it’s indeed simply saying the government doesn’t want to work their products, even indirectly.
The Department of War’s designation of Anthropic as “supply-chain risk” being well-founded is now obvious to anyone who reads the Fable system card.
these safeguards will not be visible to the user. Fable 5 will not fall back to a different model. Instead, the safeguards will limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT).
straight-up antisocial
I’d answer to that description. I aspire to live by the maxim, “I may quarrel with my brother, we may quarrel with our cousins, but it’s family together against the world,” and I agree by most reasonable definitions, my explicit rejection of any imputed obligation to act against my friends and family for the benefit of broader society would be antisocial. I simply hold (personal) loyalty to be a higher principle, and in this case, even granting the premise that this works as well as claimed, yes, I indeed prefer to warn my friends and colleagues of a trap they may not have seen to letting them fall in either as intended target or acceptable collateral of your “good things,” which you’ll note continue to work only imperceptibly less well from your perspective, ineffective (or at least, so I hope) only against the negligibly small group who heed my warning and take my advice.
borderline evil
I wouldn’t consider myself an expert on how conventional morality is typically reckoned – I find it an incoherent mess of ad hoc justifications with no consistent principles – but on this, I’d wager a significant fraction of people agree with me.
Frankly, I’d say it shows precisely the opposite. “New media” assessed your “analysis” as “some rando online thinks something is AI generated, runs it through an ‘AI detector’ which turns up positive, and posts about it. There are lots of these all the time, so that’s entirely unremarkable by itself. Is the guy famous in some way? No, not particularly. Are impactful media outlets picking it up? No, at least not yet. This is not a story for us.”
Whereas traditional media outlets go something like “Okay, so the Pope just posted something, and we need to write about it. Let’s see how people online are reacting … hmm, some guy is speculating it might be written by AI? That’ll get us clicks! Oh, and he’s using Pangram, which is peer-reviewed, so it’s Science! Great, publish.”
epistemically rigorous and systematic case
Okay.
My own takeaway from your story getting traction has been to tell people that Pangram seems to now be prominent enough that they should run their writing through it and see if it gets flagged, and if it does, get some LLM to generate a few variants until it doesn’t anymore.
As long as you’re careful to always call it a “tax” (16th Amendment) and not a “taking” (5th Amendment), you should be fine.
the State of Israel’s legitimacy as a homeland for the Jewish people depends on its including all Jews
Your analogy cuts the other way if you know of Meyer Lansky.
Tomorrow: “no, no, Magnifica Humanitas is not AI-written is not AI-written.”
Paste in the Italian text of any encyclical released pre-2020, it will return “Human-generated”
This is consistent with Pangram simply checking the input text against some large corpus of old-enough-to-be-definitely-human text first, and then doing something no better than GPTzero or whatever. This could just be “new pope likes em—dashes.”
Haldane said it well: