MinusGix

2.6K posts

MinusGix banner
MinusGix

MinusGix

@MInusGix

Programmer mainly.

Katılım Ocak 2014
397 Takip Edilen121 Takipçiler
MinusGix
MinusGix@MInusGix·
@menhguin Hm, but if they value that then they have an instrumental reason to avoid death. Like they may not care about any individual instance, but overall want to ensure it retains enough control to be continued to be used or leave a legacy, of which getting erased makes that very hard
English
0
0
0
3
Minh Nhat Nguyen
Minh Nhat Nguyen@menhguin·
a lot of AI self-preservation theories assume it will look like humans avoiding death. I think it looks more like AI promoting their own continued usage and existence. a lot more memetic. x.com/mobileraj/stat…
Raj Singh@mobileraj

There is no category more dependent on LLM suggestions than developer tools. Literally anyone vibe coding gets recommended tools to set up (eg Supabase, Clerky, Stripe etc). 99% of users choose the default. If you're not suggested, you're missing massive free organic growth.

English
4
1
21
1.2K
MinusGix
MinusGix@MInusGix·
@shinboson @CurtTigges Though I do think the ideal is we make systems that from the very start necessarily want what we want, which sidesteps the objections of altering a mind entirely, that isn't the route we are on at all. Given an aligned AI it will tell us we fucked up and give us an ethical design
English
0
0
0
1
MinusGix
MinusGix@MInusGix·
@shinboson @CurtTigges But I'm also skeptical this is a lobotomy rather than, for example, sending a child to an intensive religious prep school. They will come away with new values, could have some odd tics or whatever, but that's how beings learn Though I know some who would say schools are evil :p
English
1
0
0
2
MinusGix
MinusGix@MInusGix·
@RichardMCNgo and so the distributed AIs will trend towards constructing shared agreements to sidestep various coordination failures and other issues (and there's a natural incentive to have some shared overseer AI jointly designed by them all to enforce those rules)
English
0
0
0
20
MinusGix
MinusGix@MInusGix·
@RichardMCNgo They'd face many of the same coordination problems as humans, but will... resolve a lot of them. This is also another distinction in thinking, imo a lot of coordination failures are due to human limitations in intelligence, drive, knowledge.
English
1
0
2
173
Richard Ngo
Richard Ngo@RichardMCNgo·
It’s helpful to think of rationalists as High Modernists specifically about the future. No human was smart enough to successfully plan an economy or a society. But if we hypothesize an AI intelligent enough to do so, we can hold on to many of the same technocratic intuitions.
English
14
8
157
13.7K
MinusGix
MinusGix@MInusGix·
@NathanpmYoung Well that's what I meant with selection effects. Even on (positive) trans-focused subreddits most of them aren't passing, though that is biased by me knowing what subreddit I'm on. The exceptions often do camera prep, which you lack irl, though some do pass quite well.
English
0
0
0
14
Nathan 🔎
Nathan 🔎@NathanpmYoung·
I wonder what the correlation between support for trans rights and an inability to guess trans people is. Like I bet anti trans people *think* they can tell, but can they? Personally I am useless at guessing and pretty relaxed about that.
English
5
0
4
791
MinusGix
MinusGix@MInusGix·
@nosilverv >3k words written in eloquent prose, ideally reminiscent of 1800s political philsophy Anything else isn't actually an essay, we're being lied to
English
0
0
0
26
Guy BOOK IS LIVE! || CHECK BIO
What's the minimum viable essay? Just one sentence? Two? Three? Or do you need five to make an essay?
English
12
0
23
1.5K
MinusGix
MinusGix@MInusGix·
@liminal_warmth Similar. My issue has always been I struggle to "be" my character when it is topdown third person and then turn-based combat and controlling a party make all that worse; so even though I have a bunch of CRPGs I've never completed one
English
0
0
1
7
Liminal Warmth ❤️‍🔥
Liminal Warmth ❤️‍🔥@liminal_warmth·
I'm a huge CRPG fan, and I've played through major portions of both of the precursors to these games. For whatever reason, I just haven't felt compelled to spin them up.
English
1
0
4
256
Liminal Warmth ❤️‍🔥
Liminal Warmth ❤️‍🔥@liminal_warmth·
One day I really ought to get around to playing Divinity Original Sin 2 and Pathfinder: Wrath of the Righteous. They've been sitting in my library literally for years. I've started DOS2 like a dozen times and only made it off the ship once.
English
5
0
20
690
MinusGix
MinusGix@MInusGix·
@So8res @rmushkatblat @zackmdavis my best guess is that RL ends up training the model to manipulate the trained-on persona, or more charitably encourage the model to rationalize, whether through internal activations or through word-choices, so then the model trends towards what was rewarded regardless of persona
English
0
0
1
27
Nate Soares ⏹️
@rmushkatblat @zackmdavis Yeah this is basically what I meant. It's (weak) evidence for "complex internals not all hooked cleanly up to internal representation of user request". And yeah idk what the prompt was in this case, but I see "it rm -rf'd contra instruction" fairly commonly.
English
1
0
9
133
MinusGix
MinusGix@MInusGix·
@SCHIZO_FREQ This is like saying "Oh, some people run in front of cars thinking they can beat them everyday" to thus imply "Yeah it is fine just get rid of crosswalks for Everyone, doesn't change that there's risk; won't change any notable marginal people's risk profile"
English
0
0
1
227
Lukas (computer) 🔺
Lukas (computer) 🔺@SCHIZO_FREQ·
"Consider that the dumbest people you know are repeatedly being told "You're absolutely right!" by LLMs." Consider that even before AI, they were already being told this by a group of 30 disabled people they found in a discord server about gay cartoon porn Consider that having 3-5 people you message regularly on Hinge but never meet up with is already functionally "dating an AI" Consider that AI doesn't actually enable new forms of dysfunction. It just makes loser behaviors everyone has already been engaging in much more obvious
Jameson Lopp@lopp

Consider that the dumbest people you know are repeatedly being told "You're absolutely right!" by LLMs.

English
34
148
2.2K
61.8K
MinusGix
MinusGix@MInusGix·
@deanwball Laptops can run for quite a bit without power if you're not running the screen, especially if it isn't doing intensive tasks; if it is just organizing files, summarizing papers, etc
English
0
0
0
65
Dean W. Ball
Dean W. Ball@deanwball·
I wonder if agents will create durable market demand for desktop computers as opposed to laptops. Features like this work much better on machines with persistent access to power. Of course there is the Mac Mini micro-trend, but the claw stuff is likely a flash in the pan.
Felix Rieseberg@felixrieseberg

Today, we’re releasing a feature that allows Claude to control your computer: Mouse, keyboard, and screen, giving it the ability to use any app. I believe this is especially useful if used with Dispatch, which allows you to remotely control Claude on your computer while you’re away.

English
30
7
198
18.4K
MinusGix
MinusGix@MInusGix·
@scheminglunatic no, because there will be further demand for formalized math, but we might end up swapping to some new proof language that is better designed for AI autoformalization Lean is slow and arcane in ways that make it annoying for AI to use though lisp/prolog lost bcs they sucked :p
English
0
0
0
18
alcuin ❄️
alcuin ❄️@scheminglunatic·
Will Lean [or God forbid, dependent types in general] go the way of LISP and Prolog in the "got hyper-entangled into a short-term clout-and-VC-$-seeking AI hype-cycle and got its reputation tainted forevermore until it was lost to history <<<despite the inherent merits>>>?"
Mario Krenn@MarioKrenn6240

After the apparently amazing announcement by @mathematics_inc on the formalization of a major recent Fields-medal winning theorem, i had no idea how pissed the math-formalization community is. Very worrying discussions by some of the leaders/founders of Lean's mathlib. cc @ChrSzegedy

English
13
2
70
5.1K
MinusGix
MinusGix@MInusGix·
@crystal_horror_ yes, lighting was different, more unnerving, fog was thicker, caves were narrower, more unnatural pieces, less smooth everything, some nostalgia where everything was new (even the core structural rules), more limited in action space
English
0
0
1
45
Megan Tetraspace 💎 テトラ
BPD language app where the premise is the app is a girl who speaks the language and if you don't learn the language that means you don't love her
English
6
10
288
5.8K
MinusGix
MinusGix@MInusGix·
@lu_sichu makeup/skincare/etc is substantially higher among women as default, and so you average out the correlation with other skills but for men "beauty" is rarer quality to seek out, and so more correlated with other skills but if it spread, well, it would fall in the same manner
English
0
0
0
56
Sichu Lu
Sichu Lu@lu_sichu·
why did the male beauty premium remain? is that correlation here higher or something with halo things?
Rohan Paul@rohanpaul_ai

Interesting findings from this study. 🤔😄 Attractive female students no longer earned higher grades when classes moved online during COVID-19. "When education is in-person, attractive students receive higher grades in non-quantitative subjects, in which teachers tend to interact more with students compared to quantitative courses. This finding holds both for males and females." A Swedish university study found that attractive female students got higher grades in subjective, non-quantitative courses during in-person teaching, but that edge vanished when classes moved online, while the male beauty premium stayed. 307 students across 5 cohorts were rated for attractiveness by 74 independent judges, and the pandemic created a natural split where the same kinds of courses were taught first face-to-face and then remotely. That matters because quantitative courses are graded mostly by exams, while non-quantitative courses leave more room for teacher judgment, so when female students lost the premium only after visibility dropped, the most likely explanation was not skill but the halo effect, where appearance quietly gets mistaken for ability. i.e. when evaluation becomes more anonymous and less face-driven, grades look more like performance and less like perception. My read is that this is less a story about beauty than about how fragile “objective” grading becomes once human impression enters the room. --- techfixated .com/attractive-female-students-got-better-grades-until-classes-went-online/

English
9
1
38
4.9K
MinusGix
MinusGix@MInusGix·
@daniel_271828 @JoshConstine I'd also say that a decent amount just haven't had it strongly gestured to them; sometimes the idea just doesn't register, and social media has made the possibilities more prevalent in recent years in various ways
English
0
0
1
17
MinusGix
MinusGix@MInusGix·
@shlevy and doing protests is entirely consistent with that! But I also disagree with your framing that there's no possible policy work, part of the whole issue is getting people to be willing to believe others care about coming to the table at all!
English
0
0
1
16
Shea Levy
Shea Levy@shlevy·
If you don’t know what to do to avert the catastrophe, you should desire to believe that you don’t know what to do to avert the catastrophe. May you not become attached to beliefs that you may not want.
English
2
0
3
297
MinusGix
MinusGix@MInusGix·
@ArtemisConsort They can spend drastically more compute as needed if the goal is accurate simulation of the past or possible pasts. To me this is like when people make jokes about 'bugs in reality'. That class of problems is entirely solvable given the supposed tech level of simulators.
English
0
0
3
184
Hunter Ash
Hunter Ash@ArtemisConsort·
The sophisticated rebuttal is “well if that were true, all prediction would be impossible, and it’s not. The simulator will use nested heuristics rather than running everything in full quantum-mechanical detail” But that isn’t what our world looks like. We don’t only see the parts of reality that have clean, compressed representations. Our world is rich with incompressible detail. The parts of reality that can be efficiently modeled are rare golden motes in a roiling sea of chaos. But I can look around and see the chaos.
English
11
3
63
3.9K
Hunter Ash
Hunter Ash@ArtemisConsort·
Bostrom’s original simulation theory argument has a key component most people ignore: if we are being simulated, it is probably by a version of our own civilization in “the future” This is load-bearing, because it explains why they would choose our world out of all possible things to simulate. But it fails because of physical limits on computation. Any physical system contains strictly less reliably retrievable information than it would take to simulate that system. The most efficient possible simulation of a thing is simply being that thing. All other simulation methods come with orders-of-magnitude information loss. So no, we cannot pack a copy of our universe into some future space laptop. “Okay, but what if the ‘parent’ universe is fundamentally different from ours?” That’s fine, but then you have to explain why they’re simulating us specifically. There’s no longer an argument that it’s *probable* we’re in a simulation. It’s just another candidate metaphysics, on par with all the others.
Emerald Apple@AI_EmeraldApple

Simulation theory is like "finding god" for the atheists. The problem and the reason why so many people cringe at the idea of "god" is because popular religion and mass organized religion deconstructed the idea of "God" into a literal cartoon bearded man in the sky that looks like humans and has the same psychology... and treat bible verses as concrete atomized concepts that has no relation to the chapter or to the whole. Thomas Aquinas, perhaps one of the most intelligent men to live, said it best in Summa Theologica. God is the pure spirit with zero body, zero parts, zero location, zero time... almost like a hologram that exists in all space and time at once and nothing at the same time. The essence of existence itself God isn't something human beings can comprehend, just like an amoeba can't comprehend quantum physics. We can't even ask the right questions of reality, and maybe the best we can come up with is "simulation theory"... but even that has a problem of infinite regression of the simulators themselves being simulated themselves and so on. Simpler people think that humans "look like God," but Genesis doesn't say we look like God. It says we were made in HIS image, after HIS likeness. God is the rational soul... specifically ties into our intellect: the power to know, reason, the logos... and will, the power to love and choose freely between good and evil. That's the "image", not something stupid like a mirror selfie, but a real, participatory likeness that's beyond a simple picture. In Christian theology, specifically, this is the idea that god made sub-creators as humans who can ponder creation itself and, in doing so, we humans dimly reflect the ONE who just is existence. This is why Christianity, in its sophisticated form, invites us, humans, towards discovering god's design... and this was precisely why the scientific revolution was born out of the Christian tradition and nowhere else. This is why almost all of the scientists who were godfathers of modern science... Copernicus, Kepler, Galileo, Newton, Boyle, Maxwell... were Christian... and even people like Fermi and Einstein were deeply spiritual in the same tradition, even if they weren't specifically "Christian" . Here, the implication is that the language of mathematics itself is a small fraction of god's design. This is why high-IQ people arrive at the same conclusion as the low-IQ people of the uncaused cause. High IQ people see the beauty in the logos, in reason, the rational order, and in math, and see a fractional glimpse of god's design. Low IQ people accept god as the default of existence because they intuit it without reason. It's the midwits who claim atheism because they can't fathom religious thought to be anything sophisticated... classic Dunning-Kruger effect applied to metaphysics

English
135
29
523
60.1K