abramdemski

Karma: 21,442

abramdemski 15 Jan 2026 22:21 UTC
2 points
0
in reply to: Sodium’s comment on: Taking LLMs Seriously (As Language Models)
Yes, I think what I’ve described here shares a lot with Bengio’s program.

abramdemski 22 Dec 2025 21:20 UTC
5 points
1
on: Google seemingly solved efficient attention
The closest we can get is a little benchmark where the models are supposed to retrieve “a needle out of a haystack”. Stuff like a big story of 1 million tokens and they are supposed to retrieve a fact from it
This isn’t “the closest we can get”. Needle-in-a-haystack tests seem like a sensible starting point, but testing long-context utilization in general involves synthesis of information, EG looking at a novel or series of novels and answering reading comprehension questions. There are several benchmarks of this sort, EG:

https://epoch.ai/benchmarks/fictionlivebench
https://nyu-mll.github.io/quality/
https://www.scrolls-benchmark.com/

abramdemski 30 Nov 2025 21:36 UTC
5 points
0
in reply to: Roman Malov’s comment on: Does SI Disfavor Computationalism?
This is my inclination, but a physicalist either predicts that the phenomenology would in fact change, or perhaps asserts that you’re deluded about your phenomenal experience when you think that the experience is the same despite substrate shifts. My understanding of cube_flipper’s position is that they anticipate changes in the substrate to change the qualia.
From a physicalist’s perspective, you’re essentially making predictions based on your theory of phenomenal consciousness, and then arguing that we should already update on those predictions ahead of time, since they’re so firm. I’m personally sympathetic to this line of argument, but it obviously depends on some assumptions which need to be articulated, and which the physicalist would probably not be happy to make.

abramdemski 28 Nov 2025 20:08 UTC
3 points
0
on: abramdemski’s Shortform
Today’s Inkhaven post is an edit to yesterday’s, adding more examples of legitimacy-making characteristics, so I’m posting it in shortform so that I can link it separately:
Here are some potential legitimacy-relevant characteristics:
- The reasoning is logically valid.
- The assumptions of the argument are credible (this splits into many characteristics we can name:)
  - The assumptions are simple.
  - There are few assumptions.
  - The assumptions have very few degrees of freedom.
  - The assumptions are agreed upon by many humans.
  - The assumptions are a strong consensus in the relevant field(s).
  - The assumptions are very probable according to best existing models.
  - The assumptions have high-quality citations backing them up.
  - The assumptions have been very useful (EG productive axioms in mathematics).
  - The assumptions are not jointly contradictory.
- The reasoning is probabilistically valid.
  - The reasoning is not Dutch-bookable.
  - The reasoning is very difficult to Dutch-book (bounded cognition).
  - The reasoning is accuracy-maximizing.
  - The reasoning is a plausible extension of logic.
- The reasoning employs credible priors.
  - The prior is close to human priors, or priors humans justifiably endorse.
  - The priors have rigorous frequentist justification (EG probability of a prime number based on the prime number theorem).
  - The priors have empirical validation.
  - The priors have a maximum-entropy justification.
  - The priors are dominant over other priors one might want to use.
- The probability of the conclusion is very high.
- The argument is very strong as a statistical test.
  - The bias is low.
  - The variance is low.
  - Confidence intervals are small.
  - The test has a low false positive rate.
  - The test has a low false negative rate.
  - The test is the best test to use out of alternative tests.
  - Robustness to outliers.
  - Robustness to a variety of distributions.
  - Good rate of convergence.
What links here?
- Inkhaven Retrospective by abramdemski (30 Nov 2025 21:25 UTC; 40 points)

abramdemski 27 Nov 2025 16:51 UTC
2 points
0
in reply to: cubefox’s comment on: Infinitesimally False
Yeah, the logic still can’t handle arbitrary truth-functions; it only works for continuous truth-functions. To accept this theory, one must accept this limitation. A zealous proponent of the theory might argue that it isn’t a real loss, perhaps arguing that there isn’t really a true precise zero, that’s just a model we use to understand the semantics of the logic. What I’ll say is that this is a real compromise, just a lesser compromise than many other theories require. We can construct truth-functions arbitrarily close to a zero detector, and their corresponding Strengthened Liar will be arbitrarily close to false.

abramdemski 27 Nov 2025 2:37 UTC
7 points
1
in reply to: johnswentworth’s comment on: Courtship Confusions Post-Slutcon
Seems to me like both.

abramdemski 26 Nov 2025 22:48 UTC
18 points
11
in reply to: johnswentworth’s comment on: Courtship Confusions Post-Slutcon
No disagreement with the broad statements, but I note that your words do not particularly register the point that good conversation itself might be a turnon and lack thereof a turnoff? IE your post presents a puzzle: what’s with the banter → sex thing? I’m suggesting that many people might want to talk first as an inherent preference. Sure, there might be ways around that, but you weren’t asking for something with no loopholes, you were asking about the banter → sex thing.

abramdemski 26 Nov 2025 20:20 UTC
23 points
11
on: Courtship Confusions Post-Slutcon
Not really an experienced player of the relevant games, but I personally have turned down an obvious sex invitation with someone who I was otherwise interested in because too little conversation (and don’t regret this choice). I am not very interested in sex with someone who I can’t have a good conversation with. I feel like a lot of the intrigue of an intimate encounter is conversational intimacy. I’ve never experienced the chat at party → sex pipeline, however. Only [chat online for multiple months]->sex.

abramdemski 25 Nov 2025 22:46 UTC
LW: 28 AF: 10
18
AF
on: Alignment will happen by default. What’s next?
I also don’t believe that insider trading is immoral. Insider trading increases the accuracy of the stock prices available to the public, which is the public good that equity trading provides. For this reason, prediction markets love insider trading. The reason it’s illegal is to protect retail investors, but why do they get privileged over everyone else? Another reason insider trading is immoral is that it robs the company of proprietary information (if you weren’t a limb of The Company, you wouldn’t know the merger is happening). That’s fair, but in that case doing it officially for The Company should be allowed, and it’s not. In this example ChatGPT arguably helped steal information from LING, but did so in service of the other company, so I guess it’s kind of an immoral case—but would be moral if LING is also insider-trading on it.
The problem with insider trading, in my view, is that someone with an important role in the company can short the stock and then do something really bad that tanks the value of the company. The equilibrium in a market that allows insider trading involves draconian measures within the companies themselves to prevent this sort of behavior (or else, no multi-person ventures that can be publicly traded).
This is an instance of the more general misalignment of prediction markets: whenever there’s something on a prediction market that is quite improbable in ordinary circumstances but could be caused to happen by a single actor or a small number of people, there’s profit to be made by sewing chaos.
What links here?
- Alignment will happen by default. What’s next? by Adrià Garriga-alonso (25 Nov 2025 7:14 UTC; 99 points)

abramdemski 24 Nov 2025 16:40 UTC
2 points
0
in reply to: Mateusz Bagiński’s comment on: Informality
all the relationships between components, etc, I made “explicit”
Is there a typo here? “are made explicit” perhaps?
or that there is some set of rules such that following those rules (/modifying the expression according to those rules) is guaranteed to preserve (something like) the expression’s “truth value”?
That’s correct. More generally (since the concept also applies to noun phrases) guaranteed to preserve its “value” whatever type that may be. This “value” is something like what-it-points-at, semantic reference.

abramdemski 22 Nov 2025 15:30 UTC
5 points
0
in reply to: Steven Byrnes’s comment on: Market Logic I
Yep, will fix.

abramdemski 22 Nov 2025 5:32 UTC
2 points
0
in reply to: Gurkenglas’s comment on: Infinitesimally False
(Also it’s kinda iffy that weak disjunction is a stronger statement than strong disjunction...)
Yeah! I’m just going with what Wikipedia said there (unless I’ve made an error), but I had the same thought.

abramdemski 22 Nov 2025 5:30 UTC
2 points
0
in reply to: Archimedes’s comment on: Infinitesimally False
Fixed!

abramdemski 21 Nov 2025 17:07 UTC
2 points
1
in reply to: Gurkenglas’s comment on: Infinitesimally False
Not sure exactly what you’re getting at? I suppose you’re imagining extending the sequence to ordinal values? We can do this, sure, and still can have the same issue with sup/inf not being well-defined.

abramdemski 21 Nov 2025 17:05 UTC
2 points
0
in reply to: Gurkenglas’s comment on: Infinitesimally False
Fixed!

abramdemski 21 Nov 2025 16:55 UTC
4 points
0
in reply to: Lorxus’s comment on: Infinitesimally False
I think for Yablo’s Paradox, real truth values are adequate; you just assign ¹⁄₂ to everything.

abramdemski 20 Nov 2025 19:28 UTC
4 points
0
in reply to: Adele Lopez’s comment on: Questioning Computationalism
Sahil’s version of FGF makes many empirical predictions. In the spring of this year, we’re going to focus on listing them, so there’ll be a better answer to your question. In the context of the OP, one such prediction is that sim-Abram will have difficulty engaging in the outside world, even with some contact with real-Abram (but obviously this prediction isn’t very near-term).
With respect to your specific concern, I have a similar worry that Sahil’s FGF won’t be readily falsified by practical autonomy experiments. Full FOOM-level autonomy requires AI to replace the whole global economy; anything less will still involve “dependence on humans”.
However, it also seems plausible to me that Sahil’s FGF would make some falsifiable prediction about near-term performance on Vending-Bench 2.

abramdemski 20 Nov 2025 19:16 UTC
2 points
0
in reply to: Mateusz Bagiński’s comment on: Questioning Computationalism
Thanks, fixed!

abramdemski 18 Nov 2025 17:50 UTC
17 points
14
on: The Dutch are Working Four Days a Week
Interestingly, this is one of the issues where libertarians and progress studies people, who usually get along well, would disagree. Libertarians would say that if you can afford it, by all means, work just one day a week. Progress studies people would point out that GDP growth decreased by, say, 1% over 100 years will leave people in the resulting economy almost three times poorer.
I think this model is mistaken, and overly worships GDP as a measure of value. You’re defining non-job value-production out of existence. For example, if someone stays at home and does dishes, laundry, and raises children, this doesn’t count. If instead a nanny and a maid are hired, this counts for GDP. If someone contributes to Wikipedia in their spare time, this doesn’t count. If they’re paid to write crappy ad-copy instead, this counts for GDP. Writing a video game counts if you sell it, but doesn’t count if you give it away for free. Etc.
The libertarian view seems deeply superior here, because it trusts people’s own sense of what is valuable, rather than accepting a numerical proxy.

abramdemski 18 Nov 2025 17:26 UTC
6 points
0
in reply to: Seth Herd’s comment on: Continuity
Yeah, in my mind it is some existential risk such as the AI safety problem, and absorbing the skills of a bunch of really skilled people seems necessary to solve the problems. The thing that initially interested me here was the idea of a memory thief becoming a different person every time they absorbed someone’s memories, but somehow continuing to do it anyway. I wrote this quickly (first draft in about one hour), and wasn’t sure what the ending would be as I was writing. My initial idea was to try and write a plausible emotional arc where she started out angry at the memory thief, but as she faced the identity confusion of the situation, plausible individual steps took her to anger with her original-body-self who still had the life new-self remembered working so hard for. However, the mysterious twist ending seemed better. I wasn’t really sure if it “landed” well but two test readers said it worked for them & was better as a mystery.