Lorxus

• Lorxus 21 Nov 2025 17:10 UTC

  1 point

  0

  on: Some Hacky ELK Ideas

  Thief With Access To The Real Diamond

  That leaves a gap between cases in which the thief has access to the predictor/​planner/​reporter (which run into the diagonalization barrier), and cases in which the thief doesn’t even have detailed information about the diamond (in which case we can solve the problem). In between those, the thief has arbitrarily good information about the diamond, but does not have access to the predictor/​planner/​reporter.

  For this case, I do not have a solution, but I do expect that it’s solvable-in-principle.

  The main difficulty in these cases is that the $P\left[Obs|do\right(A^{\ast }\left)\right]$ may be exactly the same between action-plans $A^{\ast }$ which result in the diamond being stolen, and action plans which do not. The thief has enough information to make the predicted sensor readings completely identical between the two cases.

  The reason the problem might still be solvable-in-principle is that the actions $A^{\ast }$ still give us information about whether the diamond was stolen. But we somehow have to extract that information in a way which does not require any difference whatsoever between the observation-distributions in the two cases.

  I was chewing over the ELK problem this morning and had a silly idea here. What if you deliberately introduce a specific inaccuracy in the camera, or the reporter? Have some timestamp or watermark or scratch, something that has to do with the system’s perception of the diamond and which would not be known from mere access to the diamond or at least be very expensive to fake.

Lorxus Does Halfhaven: 11/​15~11/​21

• Lorxus 21 Nov 2025 14:30 UTC

  1 point

  0

  on: In­finites­i­mally False

  Thanks for the post! It makes me wonder whether hyperfinite quantification and hyperreal truth values handle Yablo-esque paradoxes too, but I haven’t thought about that much at all yet.

• Lorxus 18 Nov 2025 18:41 UTC

  1 point

  0

  in reply to: johnswentworth’s comment on: Sta­tus Is The Game Of The Losers’ Bracket

  This is an important bit of clarification! You can do some entertaining countersignaling with a nice suit jacket and an unbuttoned button-down.

On Bat­tle-Short: What, How, and Why Not To

Lorxus Does Halfhaven: 11/​08~11/​14

• Lorxus 14 Nov 2025 2:33 UTC

  7 points

  0

  on: (Fan­tasy) → (Plan­ning): A Core Men­tal Move For Agen­tic Hu­mans?

  Both of these examples show a similar mental move: (encounter a fantasy) → (figure out how to bring that fantasy into reality)...

  To me, this move feels particularly foundational and load-bearing; it’s a core part of a healthy agentic human mind. It’s the mental move...

  This resonates with something in me. Something kinda malnourished and shaky and weirdly shaped but it resonates all the same. Most of the things I like best about myself, that I’m proudest of having done or that have caused other people to be impressed with me or that nourished me to do, were things that had this pattern. I’m trying to do more of that now. Thanks for painting it so clearly.

• Lorxus 13 Nov 2025 21:14 UTC

  3 points

  0

  on: Tools for defer­ring gracefully

  I tried, but never really figured out how, to get them to understand that they were engaging in process of downloading a consensus to defer to, and why it matters that that’s what they’re doing.

  Here’s the thing, as one such person you mentored and who you counseled not to defer: you can realize you’re deferring, and recognize that it’s not particularly endorsed, and then do it anyway because all the social pressures point that way, and so do all the grantmaking pressures, and all the selection effects. It totally matters that that’s what you’re doing! But being openly non-deferential is a very good way to paint a target on your back, to take on huge costs that you would never agree to up front. Even being quietly privately non-deferential can and will show up in the kinds of research-flavored conversations you have, and will cause you to miss connections and get excluded. You’ll wind up being held to the epistemic standards of the deferrees—much higher than those of your peers; never mind that that’s not the standard most other people get held to. You’d better be willing and able to lay out your entire argument, from scratch, for anyone who asks you, or you look worse than merely stupid—you look like a waste of time.

Lorxus Fa­vors: An Ex­per­i­ment in Self-Backed Giftlike Macroe­co­nomics (+ Ex­tra Bits)

Lorxus Does Halfhaven: 11/​01~11/​07

• Lorxus 10 Nov 2025 5:48 UTC

  3 points

  4

  in reply to: GradientDissenter’s comment on: Gra­di­en­tDissen­ter’s Shortform

  So… the biggest savings here, by far, is the rent. At a guess it’s bigger than everything else combined. If you don’t have enough friends, or your friends all live in full houses, guess you’re screwed here. Hope we’re OK with the tyranny of structurelessness?

• Lorxus 9 Nov 2025 22:08 UTC

  1 point

  0

  in reply to: Ryan Kidd’s comment on: Ryan Kidd’s Shortform

  I’d like to see a breakdown by “years since doing MATS”. What’s the retention like, basically? Another breakdown I’d like to see, either for the displayed data or the years-since-MATS one—what’s the split in AI safety between “(co)founded an org”, “joined a new org”, “joined an established org/​(research/​policy) group (technical vs governance)”, and “on a grant, no real org”, along with the existing “academic” (split by core academic/​alt-academia?), “government” (which country?), and for-profit (maybe a breakdown of product type?). In any case, thanks for posting this!

• Lorxus 9 Nov 2025 0:29 UTC

  3 points

  0

  on: In­so­far As I Think LLMs “Don’t Really Un­der­stand Things”, What Do I Mean By That?

  First off, TFTP. I marked some stuff I thought was most relevant. This is helping remind me of some things I think about LLM confabulation and lack of binding/​reasoning… I don’t have my thoughts fully formed but there’s something here about global inconsistency despite local compatibility, and how that cashes out in Problems. Something a little like an inability to define a sheaf, or homology detection, or something like that? I might say more better words later about it.

• Lorxus 4 Nov 2025 17:44 UTC

  3 points

  2

  on: Re­search Reflections

  Thanks for writing this. I’ve noticed something in the same vein as well—for the last few months I’ve felt increasingly like a lot of the pieces and systems agent foundations types use have some kind of important commonality to them, though I’m not yet sure what form that could take. Condensation and natural latents and some of Francis Rhys Ward’s work in II-MAIDS; ontology mismatch and Bayes nets and imprecise probability and category theory showing up repeatedly. There’s something there to construct, but what?

• Lorxus 2 Nov 2025 1:01 UTC

  1 point

  0

  on: Lorxus’s Shortform

  I’m doing Budget Inkhaven: https://​​tiled-with-pentagons.blogspot.com/​​search/​​label/​​budget inkhaven

  Similar to my earlier writing regimen, but shorter, faster, and slightly more daring. I’ll crosspost anything I like especially well here.

• Lorxus 1 Nov 2025 22:16 UTC

  3 points

  0

  in reply to: johnswentworth’s comment on: Why Is Print­ing So Bad?

  Hm. So it’s some larger problem of which the Curse of It From Bit is a subcluster, then? If I’ve understood you right, this is also something I’ve spent time thinking about by way of finding them with my face. Something like… some unfeeling machine/​system/​structure that fails to live up to its own spec, or which has a bad spec or no spec at all, and which makes that state of affairs your problem, you silly goose who wants to actually do things, you. There’s the flavor of that thing from Zen and the Art of Motorcycle Maintenance where there’s this tiny trivial-feeling problem that humiliates you with a combination of feeling trivially small while nonetheless demanding close attention and deep understanding by way of being a critical blocker to solving whatever problem, I think? I think that the yakshaving bit is downstream of the spec failure piece—it seems like a common failure mode of degraded systems/​designs.

• Lorxus 1 Nov 2025 21:53 UTC

  18 points

  3

  on: Why Is Print­ing So Bad?

  Oh! I have thoughts about this one!

  On my model, it’s a matter of basically anything that turns Bits into Its being kinda cursed, especially if whatever process lacks human actuation or even oversight and isn’t done to truly exacting standards. Printing is an example of this; 3D printing and general CNC is another; graphic design (and the depths of what a color is) is a third. I’m thinking of the kind of thing you might find on r/​shittyrobots as well; a minimal non-example might be polypeptide/​oligonucleotide preparation, though I’m not totally clear on the workflow there or how much human labor is involved.

  My gears-level model here is most starkly illustrated by 3d printing—something about the physical object creation might change, or be underspecified, or rely on shoddily-made connectors of some kind, and in any case, the actual machine that does the actuating isn’t set up to check for failures during workflow or indeed do much more than the simple actuation and an initial calibration step. On top of that, the thing goes from being data which is everywhere and wherever you want it, to being a specific thing in a specific place; often you don’t even get to have the affordance of knowing in advance which place it is. Possibly even the yakshaving you notice is a matter of some kind of associated ugh field.

A Sketch of Helpful­ness The­ory With Equiv­o­cal Principals

• Lorxus 26 Oct 2025 22:16 UTC

  4 points

  1

  in reply to: Gurkenglas’s comment on: Seven-ish Words from My Thought-Language

  Spoons are for doing ordinary things, doing things at all. [Blood] is for doing difficult things, doing them more or harder, doing them at significant internal cost.

Seven-ish Words from My Thought-Language