For instance, GoogLeNet has multiple floppy ear detectors that appear to detect slightly different levels of droopiness, length, and surrounding context to the ears.
BTW I highly recommend you check out the post I got this #quote from – it’s about neural net interpretability, has some really cool interactive demos, and is admirably and impressively extensive. Even if you’re not confident you can understand this subject on a technical level, it’s worth checking out for the sheer coolness of the visualizations. Best Dog Hallucination Simulator of 2018, hands down
It’d be interesting to explore some of the more robust adversarial examples using these tools, like the weird turtle we were talking about a while ago
(I found this by randomly thinking, “I wonder whether Chris Olah has written any new posts about neural nets since I last looked at his blog, his neural net posts are always great,” and indeed, he had and it was great)
For instance, GoogLeNet has multiple floppy ear detectors that appear to detect slightly different levels of droopiness, length, and surrounding context to the ears.
Metaphors are, apparently, here to help, but Double Metaphors “grab hold of your true thoughts and feelings and devour them one after another, fattening themselves … They have been dwelling in the depths of your psyche since ancient times.”
While we’re on the topic of that “new Sokal hoax” stuff — IMO, the whole process of peer review and getting published is, within academia, more akin to writing job applications than doing a job. Or maybe akin to writing up some sort of progress report for a demanding but semi-clueless boss.
Academics want to get published so they can keep their jobs and advance in their careers; the actual process, including peer review (from both sides), is treated mostly as a necessary annoyance and rarely if ever as a load-bearing part of the intellectual process. (I’ve never heard anyone say “you know, reviewer #2 made a really insightful point!”, although I’m sure it happens occasionally.) And academics will eagerly step outside of this pipeline when they have the chance, e.g. with the Arxiv.
Among other things, this calls into question how much you can really immerse yourself in a research culture — as the hoaxers say they did — just by participating in its journal submission process. It’s a bit like trying to prove some point by creating a deliberately “bad” profile on LinkedIn and then trying to get referrals for from people. You might get some, but that’s not the same as (1) getting the corresponding jobs and (2) actually doing them, and behaving optimally on LinkedIn is not the most important part of most people’s job duties.
tl;dr: this latest academic journal hoax is over-hyped and the reporting on it is terrible
A trio of academics submitted 20 ridiculous papers to various feminist/gender/related-studies journals in an effort to show the journals to be ridiculous. 7 papers were accepted. The coverage has been gloating and the Twitter response has been gleeful. But the more I look into it, the less there is to it. This is troubling, because smart people like Paul Graham and Patrick Collison have retweeted about it.
WSJ articleThe Chronicle of Higher Education article
Google Drive link with all the papers and the review comments
Here’s the trio’s essay on it. At times, I think they’re deliberately vague about which ridiculous papers were accepted and which weren’t. Here’s a paragraph of theirs:
We used other methods too, like, “I wonder if that ‘progressive stack’ in the news could be written into a paper that says white males in college shouldn’t be allowed to speak in class (or have their emails answered by the instructor), and, for good measure, be asked to sit in the floor in chains so they can ‘experience reparations.’” That was our “Progressive Stack” paper. The answer seems to be yes, and feminist philosophy titan Hypatia has been surprisingly warm to it. Another tough one for us was, “I wonder if they’d publish a feminist rewrite of a chapter from Adolf Hitler’s Mein Kampf.” The answer to that question also turns out to be “yes,” given that the feminist social work journal Affilia has just accepted it.
The parallel structure of the paragraph, with ‘The answer to that question also turns out to be “yes”’ elides the very different fates of the two papers. Hypatia didn’t publish the Progressive Stack paper, and in fact they rejected it three times. But phrasing it this way, you can describe it in the same paragraph as an accepted paper, and many people won’t remember the difference. (Here’s a Harvard lecturer’s thread, with 10,000 Twitter Likes, describing the Progressive Stack paper as accepted.)
The coverage has been even worse. Here’s a Quillette piece on it, with a part that a Facebook friend quoted:
[Hypatia] invited resubmission of a paper arguing that “privileged students shouldn’t be allowed to speak in class at all and should just listen and learn in silence,” and that they would benefit from “experiential reparations” that include “sitting on the floor, wearing chains, or intentionally being spoken over.” The reviewers complained that this hoax paper took an overly compassionate stance toward the “privileged” students who would be subjected to this humiliation, and recommended that they be subjected to harsher treatment.
This isn’t just wrong; if anything, the reviewers opposed the shaming technique. Here are the full review comments for all three rejections of the paper. I don’t see any concern for an overly compassionate stance, or any recommendation of harsher treatment. When a reviewer does mention it, their concern is that it might be ineffective, and they’re uncomfortable with it. Here’s a quote from the second rejection:
What are experiential reparations? Say more about this. Also, some of your suggestions strike me as “shaming.” I’ve never had much success with shaming pedagogies, they seem to foment more resistance by members of dominant groups.
And from the same reviewer in the third rejection:
Find a place for the experiential reparations. This still makes me feel uncomfortable, because it’s shame-y and I’m not sure that student can see it otherwise.
After reading the reviewer comments, I’m very sympathetic to the reviewers, and I update toward thinking that their field is not a made-up illegible jargon-fest. They say things like:
“There are dozens of claims that are asserted and never argued for.”
“The author promises to explore key terms and explain why they are applicable to the classroom. They introduce: epistemic violence, epistemic oppression, epistemic violence, testimonial smothering, privilege-evasive epistemic pushback, epistemic exploitation, testimonial injustice, hermeneutical injustice, willful ignorance, virtuous listening, and strategic ignorance. This is too much ground to cover!!”
“The scholarship is not as sound as it could be; that is, the basic structure of the argument is plausible and interesting, but the submission has far too many issues that get in the way of a clear and sound presentation of the author’s argument.”
“I think these are basically good insights, they need to be argued for more clearly and not just asserted as true. They are interesting claims, say more, say how, say why, and don’t just assert…Explain.”
These aren’t possible comments from a field full of fashionable nonsense that doesn’t mean anything. I’m sad to contemplate the reviewers trying to help someone fix the mistakes in their paper, while the authors’ intention is to slip through as many mistakes as possible. As the editor wrote in an encouraging cover letter:
At the same time ref #1 is encouraging about your revisions. You’ll note that ref #1 says, for example, that it’s your earlier improvements that have generated some of the new problems that need attention!
See also this Twitter thread by one of the reviewers for the Masturbation is Rape paper (which was rejected). It’s sad - he rejected the paper, but wrote some encouraging things, and the hoaxers quoted the positive parts in their essay.
I haven’t looked at all the papers in detail; this isn’t a thorough investigation of all of it. Maybe I happened across the least-bad papers and the most-misleading coverage first. I think the “fat bodybuilding” paper is just as bad as it sounds: “fat bodybuilding” would be unhealthy, unpopular, and no sport has ever been started by someone proposing it in a paper to an obscure journal.
But other accepted papers, I think, use a trick: invent some fake data of interest to the journal, and include a discussion section with some silly digressions. The journal accepts the paper because the core is the interesting data, and then the hoax coverage says that the paper is about the silly digressions. For example, the core of the dog park paper is a fake observational study showing that humans, especially males, are faster to stop male-on-male dog sexual encounters than male-on-female sexual encounters. I think that’s fine; it is actually indicative of heteronormativity or homophobia or whatever. The paper also has an angle about canine rape culture, and that is indeed silly, but the paper is not best described, as The Chronicle of Higher Education did, as being “about canine rape culture in dog parks in Portland”.
There are things to learn from this whole thing. I have a lower opinion of fat studies than I did before. But I have a higher opinion of the various fields that correctly objected to ideology-pleasing buzzword-filled digressions, and I wish the coverage noted that in equal measure. I get the impression you have to fake some interesting data to get much Sokal-style fashionable nonsense through, and even then, they’ll catch most of it.
(Maybe I’m minimizing the ridiculousness of what did get past the reviewers. I think a younger, more idealistic version of me would have been more shocked by it, like the commenters at Hacker News who think that peer review should be able to detect fabricated data. My mild reaction is partly due to not expecting Idealized Science-level rigor of these fields to start with.)
And no-one should be saying anything about the rejected papers, except for praising the journals for rejecting them. If you ask someone out, and they say they’re flattered but they only like you as a friend, don’t gloat that they said that they like you. It’s a rejection.
(via kitswulf)
She mostly uses carbon dioxide based cool drink sorcery in battle.
i am extremely cute. def me writing this.
i am so pretty in the morning when my hair is damp and my eyes are a lil bit sleepy
i am the best husband
“Snakes!” he whispered. “By God, the sky is full of snakes.”
Undine hugged him spontaneously. “That was very well done! I wish I could have gotten hold of you when you were young. I could have made a wizard out of you.”
Here’s a post about Hard Problem of Consciousness, since @argumate and @foolishmeowing have talking about it lately:
I think it’s a mistake to view the Hard Problem as unique to materialism. Idealism can’t answer it either, and generally doesn’t try to. IMO, the problem is not really about matter, but about description or explanation.
(I also don’t think it’s unique to “formal” systems or approaches, except in a sense so broad that any philosophy that could ever be done is “formal,” because it involves strings of words and/or arguments.)
The Hard Problem is very similar to the problem of existence – “why is there something rather than nothing?” Both of these are questions about what “animates” or “turns on” any given description – what makes a description (such as a formal system) more than “mere words on a page.” This is a distinctive class of problem because any familiar kind of explanation would simply become part of the description, and thereby be subject to the exact same problem.
If you add some sort of “existence-maker” mechanism to your description of what exists, you’re still open to the objection that the entire description, existence-maker and all, could just as well be an inert logical structure, without the extra magic of existence. This is a pretty familiar, standard point in the context of the existence question, but in discussions about consciousness, the analogous point tends to get buried under arguments about whether or not there is more of a problem for certain kinds of description – “material” or “formal” or “functional” ones, or whatever.
It seems to me that this is a problem for descriptions, period. If you look at the various dualistic and idealistic systems that have been proposed, they tend to be, well … systems: descriptive accounts of what is supposed to exist (some or all of it mental/spiritual), along with some arguments about why we should assent to the description, but nothing inherent in them to light the flame and turn these descriptions necessarily into the realities they talk about. These systems do claim that the flame is in fact lit, but they generally treat this as self-evident via Descartes’ cogito or similar. At least one mind/spirit exists (by cogito), and here are some things it can conclude a priori about other existents – Leibniz’s various principles, McTaggart’s theory of determining correspondences, or whatever – and we’re off to the races.
These can be perfectly fine theories of what mind/spirit is, insofar as it exists, but they simply do not touch why/how it exists: you need the spark of a cogito to get things started, and the cogito doesn’t leave you any less in the dark about why there’s an existing mind (instead of there not being one). It just convinces you that there is one. And once you’ve decided to work within a frame where that is taken as given, you’ve given up on Hard Problems. These theories only “explain” the ineluctable experiencey-ness of experience in the way that the observation “as a matter of fact, something exists” explains why there is something rather than nothing – which is to say, not at all.
It seems intuitively clear to me that these Hard Problems are unanswerable, because they ask for something that is incompatible with what we take to constitute an “answer” to a question. They ask for an argument that some description is necessarily animated, that there’s no mystery about how it becomes more than words on a page because there is something impossible about the merely-words version of it. But such an argument is either:
(1) An argument for purely logical necessity, i.e. necessity within the terms of the description, in which case the necessity property is just one more fact about the description and could be as “mere” as the rest, or
(2) An argument that the description gets necessarily lit up by the animating fire of something else that already has it, in which case we need some initial spark to start things up, one that is not explained within the terms of the description. Generally this spark is supposed to be “obvious” / a priori, but the fact that we have a priori knowledge of something doesn’t constitute an explanation of why we have that knowledge, so this doesn’t get the job done.