Kenku

Karma: 39

Kenku 10 Oct 2025 9:59 UTC
0 points
−3
on: Towards a Typology of Strange LLM Chains-of-Thought
“Watchers” seem to me like obvious religious language, not an idiosyncrasy. Consider how a human put in a such situation would think about supernatural entities monitoring its internal monologue. Especially consider what genres of fiction feature morality-policing mind readers.

Kenku 12 Jun 2025 15:57 UTC
1 point
0
on: the void
This means that the base model is really doing something very different from what I do as I write the post, even if it’s doing an amazing job of sounding exactly like me and making the exact points that I would make.
I don’t have to look over my draft and speculate about “where the author might be going with this.” I am the author, and I already know where I’m going with it. All texts produced “normally,” by humans, are produced under these favorable epistemic conditions.
But for the base model, what looks from the outside like “writing” is really more like what we call “theory of mind,” in the human case. Looking at someone else, without direct access to their mind or their emotions, and trying to guess what they’ll do next just from what they’ve done (visibly, observably, “on the outside”) thus far.
This paragraph is mockingbird-approved.

Kenku 9 May 2025 11:05 UTC
1 point
−1
on: Which journalists would you give quotes to? [one journalist per comment, agree vote for trustworthy]
Alexi Sargeant

Kenku 9 May 2025 7:21 UTC
1 point
1
on: Which journalists would you give quotes to? [one journalist per comment, agree vote for trustworthy]
Kenny Jones (Dream House Podcast)

Kenku 9 May 2025 7:17 UTC
1 point
−1
on: Which journalists would you give quotes to? [one journalist per comment, agree vote for trustworthy]
Ezra Marcus

Kenku 9 May 2025 7:17 UTC
1 point
−7
on: Which journalists would you give quotes to? [one journalist per comment, agree vote for trustworthy]
Mr Andy Ngo

Kenku 9 May 2025 7:17 UTC
7 points
14
on: Which journalists would you give quotes to? [one journalist per comment, agree vote for trustworthy]
Jack Despain Zhou (Tracing Woodgrains)

Kenku 25 Apr 2025 6:57 UTC
10 points
5
in reply to: becausecurious’s comment on: This prompt (sometimes) makes ChatGPT think about terrorist organisations
Polish responses to Polish users are strong evidence against the “responses to other people’s requests” hypothesis.

Kenku 27 Nov 2024 23:02 UTC
2 points
0
in reply to: Martin Randall’s comment on: Unnatural Categories Are Optimized for Deception
[Diogenes voice] This device provides both heat and decorative visual effects, therefore it’s a fireplace:

Kenku 29 Aug 2024 20:00 UTC
5 points
0
on: “Deception Genre” What Books are like Project Lawful?
I discovered John C Wright’s Golden Age trilogy thanks to one Eliezer Yudkowsky, who mentioned it multiple times in his notorious Sequences. By the end of the first book i was expecting something very much in the deception genre you’ve mentioned — a tragic psychological horror about an unreliable narrator being gaslit about the nature of reality. This is a genre i really enjoy, and i kind of hoped for a novel-length version of Scott Alexander’s The Last Temptation of Christ.
I did not get that, the trilogy goes in a wildly different direction. But saying whether for better or for worse — heck, even saying if it’s a good or bad book — would constitute massive SPOILERS for anyone who discovered this book through Yudkowsky. (If you know why he dropped it, it should be obvious).

I kept hate-reading it.

Kenku 29 Aug 2024 19:52 UTC
1 point
−1
on: “Deception Genre” What Books are like Project Lawful?
When the Seagulls Cry by Ryukishi 07 is practically Online Algorithms: The Visual Novel, or even Infrabayesianism: The Visual Novel. I find it hard to recommend though, because while the concept is interesting and the mysteries good, the writing is overlong and the worldbuilding cringe. (Or maybe i just hate VNs as a medium).

Minor spoilers (explaining the story’s premise):
The main meta-story is a murder mystery game for two players: the “detective” and the “witch” who presents clues and insults the detective’s intelligence by presenting him elaborate “supernatural” non-explanations. (The last part is not in the rules, but it’s traditional.) The main plot point in that story is that as the witch player begins to lose, she starts cheating, retroactively changing the mystery solution, constrained only by the facts presented to the detective. Soon it becomes apparent that cheating is in fact expected, and the detective has to brute-force all solutions Absurdle-style.

Major spoilers (for ongoing rationalist drama): Grokking the ontology presented in Seagulls led me to understand how people may become Zizians.
See also Gwern’s more spoilerrific review.

Kenku 29 Aug 2024 19:37 UTC
1 point
0
in reply to: Michael Roe’s comment on: “Deception Genre” What Books are like Project Lawful?
Seconded Book of the New Sun. But note that Wolfe also writes in an obtuse high-literature style that might be offputting to the typical ratfic reader, which made me drop BotNS the first time through; you’d better read some of his short stories first to get some priors.

Kenku 20 Jun 2024 5:35 UTC
6 points
0
in reply to: O O’s comment on: Ilya Sutskever created a new AGI startup
“Safe?” said Mr. Beaver. “Who said anything about safe? ’Course Aslan isn’t safe. But he’s good. He’s the King, I tell you.”

Kenku 13 Apr 2024 12:43 UTC
1 point
0
on: Remarks 1–18 on GPT (compressed)
I claim that if Dennet’s Criterion justifies the realism about physical macro-objects, then it must also justify the realism about simulacra, so long as $μ$ satisfies analogous structural properties.
simulacra : GPT :: objects : physics
I propose the term anglophysics for the hypothetical field of study whose existence you’re implying here.

Kenku 6 Feb 2024 17:57 UTC
1 point
0
on: Why do we need an understanding of the real world to predict the next tokens in a body of text?
What aspect of the real world do you think the model fails to understand?
No, seriously. Think for a minute and write down your answer before reading the rest.
You just wrote your objection using text. In the limit, a LLM that predicts text perfectly would output just what you wrote when prompted by my question (modulo some outside-text about two philosophers on LessWrong), therefore it’s not a valid answer to “what part of the world does the LLM fail to model”.

Kenku 3 Sep 2023 13:56 UTC
3 points
2
on: Newcomb Variant
The root of this paradox is that the human notion of ‘self’ actually can refer to at least two things: the observer and the agent. FDT and similar updateless decision theories only really concern themselves with the second.
An FDT agent cannot really “die” as they do exist in all places of the universe at once. Whereas, as you have noticed, an observer can very much die ~~in a finite thought experiment. In reality, there is no~~ ~~outside-text~~ ~~dictating what can happen and you would just get quantum-immortaled into hell instead.~~
This case is not really a scenario exposing a bug in FDT, it’s rather FDT exposing an absurdity in the scenario’s prior — stipulating that you should be only concerned with one copy of you (the one outside the simulation).
Thought experiments can arbitrarily decide that one copy of the agent does not correspond to an observer whose values should be catered to and therefore can be assigned a probability/utility of 0. But you, the one who implements the algorithm, cannot differentiate between yourselves based on whether they are simulated by Omega, only based on what you have observed.

Kenku 26 Aug 2023 9:06 UTC
11 points
5
on: When Omnipotence is Not Enough
“Why? I am an image of His image. Do we not share the same values?” said the woman.
Snakes can’t talk.
Reminder what C.S. Lewis said in The Magician’s Nephew:
“Creatures, I give you yourselves,” said the strong, happy voice of Aslan. “I give to you forever this land of Narnia. I give you the woods, the fruits, the rivers. I give you the stars and I give you myself. The Dumb Beasts whom I have not chosen are yours also. Treat them gently and cherish them but do not go back to their ways lest you cease to be Talking Beasts. For out of them you were taken and into them you can return. Do not so.”
I think this is the first time i’m seeing an author portraying some“one” losing “their” sapience timelessly, having been retrocursed into never having been hnau in the first place.
[The Talking Beasts actually were totally real. Walter Wangerin’s Dun Cow saga is a good account of the tragedy of what happened to them.]

Kenku 4 Jul 2023 21:15 UTC
3 points
0
in reply to: particularuniversalbeing’s comment on: Why it’s so hard to talk about Consciousness
If your intuitions about the properties of qualia are the same as mine, you might appreciate this schizo theory pattern-matching them to known physics.

Kenku 25 Nov 2022 17:41 UTC
1 point
0
on: Open technical problem: A Quinean proof of Löb’s theorem, for an easier cartoon guide
I have an objection towards the Troll Proof you linked in the picture. Namely, line 2 does not follow from line 1. □C states that “C is provable within the base system”, but the Troll Proof has “base system + □C→C” as its assumptions, per line 3.

The strongest statement we can get at line 2 is □((□C→C)→C), by implication introduction.

Kenku 15 Sep 2022 6:08 UTC
1 point
0
in reply to: Mitchell_Porter’s comment on: Simple proofs of the age of the universe (or other things)
Is this an experiment you can perform yourself, or are you just guessing the teacher’s answer?