blakestacey @ blakestacey @awful.systems Posts 33Comments 630Joined 2 yr. ago
First reaction: "Wait, that was in Nature?"
Second reaction: "Oh, Nature Scientific Reports. The 'we have Nature at home' of science journals."
Among many insights, Davis (politely) points out that one of the AI-generated Chaucer poems is just "the opening of the Prologue to the Canterbury Tales."
Whan that Aprille with the fuck?
Throwback Thursday: Atlas Shrugged: The Cobra Commander Dialogues
(Based on blog posts now available here.)
From page 17:
Rather than encouraging critical thinking, in core EA the injunction to take unusual ideas seriously means taking one very specific set of unusual ideas seriously, and then providing increasingly convoluted philosophical justifications for why those particular ideas matter most.
ding ding ding
Abstract: This paper presents some of the initial empirical findings from a larger forth-coming study about Effective Altruism (EA). The purpose of presenting these findings disarticulated from the main study is to address a common misunderstanding in the public and academic consciousness about EA, recently pushed to the fore with the publication of EA movement co-founder Will MacAskill’s latest book, What We Owe the Future (WWOTF). Most people in the general public, media, and academia believe EA focuses on reducing global poverty through effective giving, and are struggling to understand EA’s seemingly sudden embrace of ‘longtermism’, futurism, artificial intelligence (AI), biotechnology, and ‘x-risk’ reduction. However, this agenda has been present in EA since its inception, where it was hidden in plain sight. From the very beginning, EA discourse operated on two levels, one for the general public and new recruits (focused on global poverty) and one for the core EA community (focused on the transhumanist agenda articulated by Nick Bostrom, Eliezer Yudkowsky, and others, centered on AI-safety/x-risk, now lumped under the banner of ‘longtermism’). The article’s aim is narrowly focused onpresenting rich qualitative data to make legible the distinction between public-facing EA and core EA.
From the linked Andrew Molitor item:
Why Extropic insists on talking about thermodynamics at all is a mystery, especially since “thermodynamic computing” is an established term that means something quite different from what Extropic is trying to do. This is one of several red flags.
I have a feeling this is related to wanking about physics in the e/acc holy gospels. They invoke thermodynamics the way that people trying to sell you healing crystals for your chakras invoke quantum mechanics.
They take a theory that is supposed to be about updating one's beliefs in the face of new evidence, and they use it as an excuse to never change what they think.
The existence of a Wikipedia page for dinosaur erotica must prove that back in the days when humans co-existed with stegosaurs, the ones who fucked them lived better.
Erin go Bleagh.
"Consider it from the perspective of someone who does not exist and therefore has no preferences. Who would they pick?"
The idea that formalist experimentation and deliberately pushing the boundaries of a medium are only one of several goals to which art can strive is apparently too sophisticated for Scott Adderall. He also takes a leap from "influential" to "meaningful", an elision so hackish it's trite.
It's a very "steampunk (derogatory)" picture, like something I would have found on sale in the artist room of the science-fiction convention that convinced me I don't like science-fiction conventions very much.
I think that in this particular instance, it's OK to kinkshame
"So, professor sir, are you OK with psychologically torturing Black people, or do you just not care?"
Superficially, it looks like he's making a testable prediction. But that "prediction" is a number from a bullshit calculation (or maybe two or three different, mutually inconsistent bullshit calculations — it's hard to be sure). So if someone wasted their time and did the experiment, he'd handwave away the null result by fiddling the input bullshit.
I will try to have some more comments about the physics when I have time and energy. In the meanwhile:
Entropy in thermodynamics is not actually a hard concept. It's the ratio of the size of a heat flow to the temperature at which that flow is happening. (So, joules per kelvin, if you're using SI units.) See episodes 46 and 47 of The Mechanical Universe for the old-school PBS treatment of the story. The last time I taught thermodynamics for undergraduates, we used Finn's Thermal Physics, for the sophisticated reason that the previous professor used Finn's Thermal Physics.
Entropy in information theory is also not actually that hard of a concept. It's a numerical measure of how spread-out a probability distribution is.
It's relating the two meanings that is tricky and subtle. The big picture is something like this: A microstate is a complete specification of the positions and momenta of all the pieces of a system. We can consider a probability distribution over all the possible microstates, and then do information theory to that. This bridges the two definitions, if we are very careful about it. One thing that trips people up (particularly if they got poisoned by pop-science oversimplifications about "disorder" first) is forgetting the momentum part. We have to consider probabilities, not just for where the pieces are, but also for how they are moving. I suspect that this is among Vopson's many problems. Either he doesn't get it, or he's not capable of writing clearly enough to explain it.
So these two were published in American Institute of Physics Advances, which looks like a serious journal about physics. Does anyone know about it? It occupies a space where I can’t easily find any obvious issues, but I also can’t find anyone saying “ye this is legit”. It claims to be peer-reviewed, and at least isn’t just a place where you dump a PDF and get a DOI in return.
I have never heard of anything important being published there. I think it's the kind of journal where one submits a paper after it has been rejected by one's first and second (and possibly third) choices.
However, after skimming, I can at least say that it doesn’t seem outlandish?
Oh, it's worse than "outlandish". It's nonsensical. He's basically operating at a level of "there's an E in this formula and an E in this other formula, so I will set them equal and declare it revolutionary new physics".
Here's a passage from the second paragraph of the 2023 paper:
The physical entropy of a given system is a measure of all its possible physical microstates compatible with the macrostate, SPhys. This is a characteristic of the non-information bearing microstates within the system. Assuming the same system, and assuming that one is able to create N information states within the same physical system (for example, by writing digital bits in it), the effect of creating a number of N information states is to form N additional information microstates superimposed onto the existing physical microstates. These additional microstates are information bearing states, and the additional entropy associated with them is called the entropy of information, SInfo. We can now define the total entropy of the system as the sum of the initial physical entropy and the newly created entropy of information, Stot = SPhys + SInfo, showing that the information creation increases the entropy of a given system.
wat
Storing a message in a system doesn't make new microstates. How could it? You're just rearranging the pieces to spell out a message — selecting those microstates that are consistent with that message. Choosing from a list of available options doesn't magically add new options to the list.
The "simulation hypothesis" is an ego flex for men who want God to look like them.
I sneered that in a blog post last year, as it happens.
From the Wired story:
As a comparison, Cui cited another analysis that GPTZero ran on Wikipedia earlier this year, which estimated that around one in 20 articles on the site are likely AI-generated—about half the frequency of the posts GPTZero looked at on Substack.
That should be one in 20 new articles, per the story they cite, which is ultimately based on arXiv:2410.08044.
David Skilling, a sports agency CEO who runs the popular soccer newsletter Original Football (over 630,000 subscribers), told WIRED he sees AI as a substitute editor. “I proudly use modern tools for productivity in my businesses,” says Skilling.
Babe wake up, a new insufferable prick just dropped.
Edit to add: There's an interesting example here of a dubious claim being laundered into truthiness. That arXiv preprint says this in the conclusion section.
Shao et al. (2024) have even designed a retrieval-based LLM workflow for writing Wikipedia-like articles and gathered perspectives from experienced Wikipedia editors on using it—the editors unanimously agreed that it would be helpful in their pre-writing stage.
But if we dig up arXiv:2402.14207, we find that the "unanimous" agreement depends upon lumping together "somewhat" and "strongly agree" on their Likert scale. Moreover, this grand claim rests upon a survey of a grand total of ten people. Ten people, we hasten to add, who agreed to the study in the first place, practically guaranteeing a response bias against those Wikipedians who find "AI" morally repugnant.
shocked, shocked to find that gambling is going on in here.gif