Shankar Sivarajan

Karma: 1,409

Shankar Sivarajan 10 Jun 2026 22:51 UTC
3 points
−3
on: Shankar Sivarajan’s Shortform
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).

Shankar Sivarajan 4 Jun 2026 1:40 UTC
3 points
0
in reply to: Linch’s comment on: Linch’s Shortform
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.

Shankar Sivarajan 3 Jun 2026 1:35 UTC
−11 points
−19
in reply to: Linch’s comment on: Linch’s Shortform
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.

Shankar Sivarajan 1 Jun 2026 17:05 UTC
−2 points
−1
in reply to: AnthonyC’s comment on: NYT: Senator Sanders Proposes Gov’t Take 50% Ownership of AI Labs
As long as you’re careful to always call it a “tax” (16th Amendment) and not a “taking” (5th Amendment), you should be fine.

Shankar Sivarajan 31 May 2026 0:08 UTC
1 point
0
on: Comment on “Banning Said Achmiz”
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.

Shankar Sivarajan 27 May 2026 20:01 UTC
4 points
0
on: no, Magnifica Humanitas is not AI-written
Tomorrow: “no, no, Magnifica Humanitas is not AI-written is not AI-written.”

Shankar Sivarajan 26 May 2026 13:43 UTC
14 points
1
on: Many portions of Magnifica Humanitas appear to be 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.”

Shankar Sivarajan 21 May 2026 18:53 UTC
3 points
1
in reply to: romeostevensit’s comment on: Women should be able to open things
Yeah, the technique advocated by a lot of people here seems annoying. I have a simpler one: tap the rim of the lid with the spoon/knife until it crumples a little. Like so.

Shankar Sivarajan 21 May 2026 16:11 UTC
4 points
1
in reply to: Raphael Roche’s comment on: Women should be able to open things
only too happy to watch him fail so they can have a go themselves
He loosened it for them.

Shankar Sivarajan 20 May 2026 4:48 UTC
2 points
0
in reply to: CronoDAS’s comment on: CronoDAS’s Shortform
Us physicists never beating the allegations!
Just give them Griffiths, and you’ll do okay for teaching Quantum Mechanics, but for chemistry … you’re correct, of course, that the Schrödinger equation is what all those heuristics about orbital hybridization are trying to approximate, and sure, you could add some lines emphasizing that at the beginning if it’s not clear, but I don’t think the pedagogy would be improved by dispensing with the heuristics altogether.

Shankar Sivarajan 20 May 2026 0:37 UTC
4 points
0
in reply to: CronoDAS’s comment on: CronoDAS’s Shortform
the reason the explanation made no sense was that it was entirely bullshit
the usual route of describing the history of quantum mechanics and mistaking that for an actual explanation.
Are you sure you understand Pauling’s work on the nature of the chemical bond?

Shankar Sivarajan 18 May 2026 20:17 UTC
−6 points
0
in reply to: habryka’s comment on: leogao’s Shortform
Fatal Familial Insomnia that has been modified to be contagious.
I seem to have misjudged how well-known prion diseases were. Would it help to consider an analogy to the phrase, “airborne rabies”?

Shankar Sivarajan 16 May 2026 19:23 UTC
−19 points
−2
in reply to: RHollerith’s comment on: leogao’s Shortform
The hypothetical personal reasons would have to be compelling, yes, of course. This would not be done on a whim. You will note the disease I chose as my example was heritable – even if you hadn’t heard of it before, the name should have made that clear – which lends itself readily to obvious motivations.

Shankar Sivarajan 16 May 2026 19:15 UTC
−47 points
−65
in reply to: leogao’s comment on: leogao’s Shortform
You jest, but I hold this class of strategy to be perfectly sound in principle and grossly underexplored in practice. I call it “countersteering for the cure,” and the specific example I had in mind was contagious Fatal Familial Insomnia. The idea is that if you wanted some rare/orphan disease cured for personal reasons, but lacked the resources to achieve that directly, you might be able to go in the opposite direction, and make it much more common instead, enough to trigger several orders of magnitude more funding/talent towards the cure you want.

Shankar Sivarajan 11 May 2026 2:11 UTC
2 points
0
in reply to: winwonce’s comment on: Intuition for 1 + 2 + 3 + … = −1/12
Could you give me an example of such an x ≠ 0 for which this is true?

Shankar Sivarajan 10 May 2026 0:37 UTC
−2 points
−2
on: International Law Cannot Prevent Extinction Either
powerful countries disregard international law whenever they want. A stark example of this is the Budapest Memorandum
Who, after all, today speaks of the American sanctions against Belarus?
You choosing this as an example but completely ignoring the far better illustration of this makes your point that there are absolutely no consequences to flouting such agreements if you’re powerful enough better than the rest of your post.

Shankar Sivarajan 8 May 2026 18:41 UTC
2 points
0
in reply to: AprilSR’s comment on: Matrice Jacobine’s Shortform
That’s the resolution criterion for the Manifold Market (well, technically for the Polymarket market that the Manifold one is to resolve identically to).

Shankar Sivarajan 6 May 2026 22:13 UTC
8 points
2
in reply to: DirectedEvolution’s comment on: Matrice Jacobine’s Shortform
based on a 50% drop in pandemic probability
Relevant xkcd: 1252

Shankar Sivarajan 3 May 2026 2:44 UTC
7 points
0
in reply to: Ustice’s comment on: Games that change your mind
You must have played Outer Wilds, right? If by some chance you haven’t, I highly recommend it: it seems like it’s a perfect example of the genre you like.

Shankar Sivarajan 2 May 2026 14:38 UTC
7 points
0
in reply to: gwern’s comment on: How Go Players Disempower Themselves to AI
I guess this is karma for me ever having replied to a question with a link to lmgtfy^[1].
I thought it would be clear from context that what I was asking for was a first-hand account of how (and whether) such adversarial strategies, which I read are simple enough to be possible to learn and implement unaided over-the-board, had impacted play in these no-stakes Go schools.
1. ^
  In my defense, that was like fifteen years ago, back when Google still reliably answered questions.