Here’s another possible explanation: The models aren’t actually as impressive as they’re made out to be. For example, take DallE2. Yes, it can create amazingly realistic depictions of noun phrases automatically. But can it give you a stylistically coherent drawing based on a paragraph of text? Probably not. Can it draw the same character in three separate scenarios? No, it cannot.
DallE2 basically lifts the floor of quality for what you can get for free. But anyone who actually wants or needs the things you can get from a human artist cannot yet get it from an AI.
DallE2 is bad at prepositional phrases (above, inside) and negation. It can understand some sentence structure, but not reliably.
In the first example, none of those are paragraphs longer than a single sentence.
In the first example, the images are not stylistically coherent! The bees are illustrated inconsistently from picture to picture. They look like they were drawn by different people working off of similar prompts and with similar materials.
The variational feature is not what I’m talking about; I mean something like “Draw a dragon sleeping on a pile of gold, working in a supermarket, and going to a tea party with a unicorn, with the same dragon in each image”.
DallE2 is bad at prepositional phrases (above, inside) and negation. It can understand some sentence structure, but not reliably.
Goalpost moving. DALL-E 2 can generate samples matching lots of complex descriptions which are not ‘noun phrases’, and GLIDE is even better at it (also covered in the paper). You said it can’t. It can. Even narrowly, your claim is poorly supported, and for the broader discussion this is in the context of, misleading. You also have not provided any sources or general reasons for this sweeping assertion to be true, or for the broader implications you claimed these are good support for.
In the first example, none of those are paragraphs longer than a single sentence.
What happened to ‘noun phrases’?
In the first example, the images are not stylistically coherent! The bees are illustrated inconsistently from picture to picture. They look like they were drawn by different people working off of similar prompts and with similar materials.
Those images are stylistically coherent in being clearly in a pastel style and matching the text input. That meets the demand, and this is only a quick throwaway project establishing a lower bound on what DALL-E 2 can do. “Attacks only get better.”
That they are not, in addition to this, perfectly consistent with each other is too bad, but increased similarity is well within the scope of a DALL-E architecture through, just off the top of my head, the variations functionality, direct optimization by backprop, or CLIP rejection sampling.
You also have not provided any sources or general reasons for this sweeping assertion to be true.
The variational feature is not what I’m talking about
I don’t know why you look at that and say it’s not.
You also have not provided any sources or general reasons for this sweeping assertion to be true.
This is an absurd moderation policy to use on a post which you wrote for the purpose of learning something. It strongly signals that you care much less about learning anything than you do about fussy politeness norms.
It isn’t absurd, and it doesn’t at all signal what you say it signals. Alex_Altair didn’t ask for the comment to be removed, but for it to be changed to something more kind. Learning something is important, kindness is as well.
Given that Alex not just asked, but in fact did remove gwern’s comment, it would seem that you’re mistaken. It turns out that learning something is not important to the OP, after all.
I should have checked that, I’ll admit. Still, the comment wasn’t immediately removed and gwern did get the chance to change it (and still can create a new, more polite version btw).
Also, learning something from a comment and deleting it aren’t mutually exclusive.
If, hypothetically, I were to make a post asking a question and soliciting information, and then someone replied to my question-post with a comment that contained a large amount of exactly the sort of information I had asked for, but then, rather than thanking that person for taking the time out of their day to contribute their knowledge and helping me, and anyone else reading the post, to learn precisely the things I was ostensibly trying to learn, I instead chastised the respondent for some slightly abrasive language and demanded[1] that they take more time to go back and edit their post to conform to my exacting standards of politeness…
… well, I find this scenario embarrassing even to contemplate. I hope that I never display such a degree of intellectual arrogance and close-mindedness.
As for then deleting the response in question, that is so egregiously foolish, petulant, and petty that I can’t even imagine doing it. Are we to believe that because a highly informative comment contains a single mildly sarcastic remark, it would therefore be better that the members of Less Wrong not even be exposed to it? Or are you suggesting that OP read the comment, learned from it, and then deleted it because he had gotten all the value he could from it, and as for the rest of us, we can go hang?
Still, the comment wasn’t immediately removed and gwern did get the chance to change it (and still can create a new, more polite version btw).
Why in the world would gwern want to do this? Do you think that writing comments on OP’s posts is such a singular privilege that obviously gwern is going to take the time to carefully rewrite his comments to OP’s standards? Is there any reason, at all, why the response to this sort of treatment should be anything other than a shrug and walking away? Do you think that allowing gwern (or anyone else) to comment on his posts is a favor that OP is doing him?
Are we to believe that because a highly informative comment contains a single mildly sarcastic remark, it would therefore be better that the members of Less Wrong not even be exposed to it? Or are you suggesting that OP read the comment, learned from it, and then deleted it because he had gotten all the value he could from it, and as for the rest of us, we can go hang?
I hadn’t thought of it like this. You’re right, deleting the comment was a bad action.
Why in the world would gwern want to do this?
I would want to, or at least hope I would. Not that that is a reason why gwern should want it too, but it shines some light on why I think others may want to.
Do you think that allowing gwern (or anyone else) to comment on his posts is a favor that OP is doing him?
Good point. No, but kindness is still important. (Note I don’t think gwern was that unkind. I just think Alex had a point, and think your reaction to Alex was unfair.)
It seems we have reached partial agreement though; thank you for your time.
Meta: I disagree with Alex’s decision to delete Gwern’s comment on this answer. People can reasonably disagree about the optimal balance between ‘more dickish’ (leaves more room for candor, bluntness, and playfulness in discussions) and ‘less dickish’ (encourages calm and a focus on content) in an intellectual community. And on LW, relatively high-karma users like Alex are allowed to moderate discussion of their posts, so Alex is free to promote the balance he thinks is best here.
But regardless of where you fall on that spectrum, I think LW should have a soft norm that substance trumps style, content is king, argument will be taken seriously on its own terms even if it’s not optimally packaged and uses the wrong shibboleths or whatever.
Deleting substantive, relevant content entirely should mostly not be one of the ‘game moves’ people use in advancing their side of the Dickishness Debate—it’s not worth it on its own terms, it’s not worth it as a punishment for the other side (even if the other side is in fact wrong and you’re right), and it erodes an important thing about LW.
Gwern’s comment had tons of content beyond that one sentence that was phrased a bit rudely; and it spawned a bunch of discussion that’s now hard to follow, on a topic that actually matters. Deleting the whole comment, without copy-pasting all or most of it anywhere, seems bad to me.
I’m interested in responding to you, Rob, because I already know you to be an entirely reasonable person, and also because I think this is somewhat of a continuation of a difference between you and me in real life. I might bail at any time though, because the fact that posters can have their own custom moderation policy means that I don’t feel particularly obligated to justify myself.
(For context for the rest of this comment, the line I had a problem with was, “‘noun phrases’ is an odd typo for ‘sentences’. They’re not even close to each other on the keyboard.”)
But regardless of where you fall on that spectrum, I think LW should have a soft norm that substance trumps style, content is king, argument will be taken seriously on its own terms even if it’s not optimally packaged and uses the wrong shibboleths or whatever.
I agree with this, and I think it’s already true. But I also think you worded it too softly to be in contradiction to my comment deletion (and more generally the implicit policy in my head). LW definitely does have said soft norm; I think allowing users to moderate their own posts, and users occasionally doing so, preserves that norm! Never deleting a comment, no matter where it lay on the dickish spectrum, would I think constitute a hard norm. The line I had a problem with was far beyond “not optimally packaged” and had nothing to do with using a “wrong shibboleths”.
Deleting substantive, relevant content entirely should mostly not be one of the ‘game moves’ people use in advancing their side of the Dickishness Debate—it’s not worth it on its own terms, it’s not worth it as a punishment for the other side (even if the other side is in fact wrong and you’re right), and it erodes an important thing about LW.
I’ll note that I gave them plenty of time and opportunity to edit the comment; I requested it in the comments, I requested it in a PM, and I saw that they made other comments on the same post much later.
and it spawned a bunch of discussion that’s now hard to follow, on a topic that actually matters. Deleting the whole comment, without copy-pasting all or most of it anywhere, seems bad to me.
It spawned two comments (which you can still read, though I agree they’re harder to follow). I agree that the rest of said content was substantive and relevant. If there had actually been a whole lot more, I probably wouldn’t have deleted the parent. I just don’t think it was enough to tip the scale. And there’s tons of other similar discussion all over the comments, especially by Gwern. And like, everyone involved is free to reiterate the substantive and relevant content. (The option of copy-pasting someone’s comment feels weird to me, for reasons that I haven’t quite explicated and don’t feel super relevant.)
Gwern’s comment had tons of content beyond that one sentence that was phrased a bit rudely
It really wasn’t that big of a comment. But also, I think our core disagreement might be here under “phrased a bit rudely”. I think the criterion I’m using implicitly here is that, if the entire purpose of the statement is to insult someone, then it’s out. That sentence did not have any other purpose. It’s not just phrasing that comes a bit too harsh. It was equivalent to, “You’re an idiot”. This is far from the side of the spectrum that requires finicky social norms. I’m not asking that people make sure that no one could be offended by what they’re saying. I’m not even saying that people shouldn’t say true, useful negative things about others. If it’s important to figuring out AI alignment that we discuss how smart a particular person is, then so be it. That’s a far cry from statements whose only purpose is to insult.
I’d actually be curious to know where the line is for you. If someone literally said, “You’re an idiot”, would you call that too dickish? Or what if someone’s comment had insulting profanities?
(These are all quantitative factors. If Gwern’s overall comment had sucked more, or his sentence had been way more egregious, I’d have objected a lot less to Alex’s call. But it does matter where we put rough quantitative thresholds.)
Here’s another possible explanation: The models aren’t actually as impressive as they’re made out to be. For example, take DallE2. Yes, it can create amazingly realistic depictions of noun phrases automatically. But can it give you a stylistically coherent drawing based on a paragraph of text? Probably not. Can it draw the same character in three separate scenarios? No, it cannot.
DallE2 basically lifts the floor of quality for what you can get for free. But anyone who actually wants or needs the things you can get from a human artist cannot yet get it from an AI.
See also, this review of a startup that tries to do data extraction from papers: https://twitter.com/s_r_constantin/status/1518215876201250816
DallE2 is bad at prepositional phrases (above, inside) and negation. It can understand some sentence structure, but not reliably.
In the first example, none of those are paragraphs longer than a single sentence.
In the first example, the images are not stylistically coherent! The bees are illustrated inconsistently from picture to picture. They look like they were drawn by different people working off of similar prompts and with similar materials.
The variational feature is not what I’m talking about; I mean something like “Draw a dragon sleeping on a pile of gold, working in a supermarket, and going to a tea party with a unicorn, with the same dragon in each image”.
Goalpost moving. DALL-E 2 can generate samples matching lots of complex descriptions which are not ‘noun phrases’, and GLIDE is even better at it (also covered in the paper). You said it can’t. It can. Even narrowly, your claim is poorly supported, and for the broader discussion this is in the context of, misleading. You also have not provided any sources or general reasons for this sweeping assertion to be true, or for the broader implications you claimed these are good support for.
What happened to ‘noun phrases’?
Those images are stylistically coherent in being clearly in a pastel style and matching the text input. That meets the demand, and this is only a quick throwaway project establishing a lower bound on what DALL-E 2 can do. “Attacks only get better.”
That they are not, in addition to this, perfectly consistent with each other is too bad, but increased similarity is well within the scope of a DALL-E architecture through, just off the top of my head, the variations functionality, direct optimization by backprop, or CLIP rejection sampling.
You also have not provided any sources or general reasons for this sweeping assertion to be true.
I don’t know why you look at that and say it’s not.
You also have not provided any sources or general reasons for this sweeping assertion to be true.
Don’t be a dick. As a moderator of my own post, I request that you change this to not be insulting.
This is an absurd moderation policy to use on a post which you wrote for the purpose of learning something. It strongly signals that you care much less about learning anything than you do about fussy politeness norms.
It isn’t absurd, and it doesn’t at all signal what you say it signals. Alex_Altair didn’t ask for the comment to be removed, but for it to be changed to something more kind. Learning something is important, kindness is as well.
Given that Alex not just asked, but in fact did remove gwern’s comment, it would seem that you’re mistaken. It turns out that learning something is not important to the OP, after all.
I should have checked that, I’ll admit. Still, the comment wasn’t immediately removed and gwern did get the chance to change it (and still can create a new, more polite version btw).
Also, learning something from a comment and deleting it aren’t mutually exclusive.
If, hypothetically, I were to make a post asking a question and soliciting information, and then someone replied to my question-post with a comment that contained a large amount of exactly the sort of information I had asked for, but then, rather than thanking that person for taking the time out of their day to contribute their knowledge and helping me, and anyone else reading the post, to learn precisely the things I was ostensibly trying to learn, I instead chastised the respondent for some slightly abrasive language and demanded[1] that they take more time to go back and edit their post to conform to my exacting standards of politeness…
… well, I find this scenario embarrassing even to contemplate. I hope that I never display such a degree of intellectual arrogance and close-mindedness.
As for then deleting the response in question, that is so egregiously foolish, petulant, and petty that I can’t even imagine doing it. Are we to believe that because a highly informative comment contains a single mildly sarcastic remark, it would therefore be better that the members of Less Wrong not even be exposed to it? Or are you suggesting that OP read the comment, learned from it, and then deleted it because he had gotten all the value he could from it, and as for the rest of us, we can go hang?
Why in the world would gwern want to do this? Do you think that writing comments on OP’s posts is such a singular privilege that obviously gwern is going to take the time to carefully rewrite his comments to OP’s standards? Is there any reason, at all, why the response to this sort of treatment should be anything other than a shrug and walking away? Do you think that allowing gwern (or anyone else) to comment on his posts is a favor that OP is doing him?
Phrasing the demand as a “request”, prior to enforcing said demand, does not actually make it a request—merely a lie as well.
I hadn’t thought of it like this. You’re right, deleting the comment was a bad action.
I would want to, or at least hope I would. Not that that is a reason why gwern should want it too, but it shines some light on why I think others may want to.
Good point. No, but kindness is still important. (Note I don’t think gwern was that unkind. I just think Alex had a point, and think your reaction to Alex was unfair.)
It seems we have reached partial agreement though; thank you for your time.
Meta: I disagree with Alex’s decision to delete Gwern’s comment on this answer. People can reasonably disagree about the optimal balance between ‘more dickish’ (leaves more room for candor, bluntness, and playfulness in discussions) and ‘less dickish’ (encourages calm and a focus on content) in an intellectual community. And on LW, relatively high-karma users like Alex are allowed to moderate discussion of their posts, so Alex is free to promote the balance he thinks is best here.
But regardless of where you fall on that spectrum, I think LW should have a soft norm that substance trumps style, content is king, argument will be taken seriously on its own terms even if it’s not optimally packaged and uses the wrong shibboleths or whatever.
Deleting substantive, relevant content entirely should mostly not be one of the ‘game moves’ people use in advancing their side of the Dickishness Debate—it’s not worth it on its own terms, it’s not worth it as a punishment for the other side (even if the other side is in fact wrong and you’re right), and it erodes an important thing about LW.
Gwern’s comment had tons of content beyond that one sentence that was phrased a bit rudely; and it spawned a bunch of discussion that’s now hard to follow, on a topic that actually matters. Deleting the whole comment, without copy-pasting all or most of it anywhere, seems bad to me.
I appreciate this comment!
I’m interested in responding to you, Rob, because I already know you to be an entirely reasonable person, and also because I think this is somewhat of a continuation of a difference between you and me in real life. I might bail at any time though, because the fact that posters can have their own custom moderation policy means that I don’t feel particularly obligated to justify myself.
(For context for the rest of this comment, the line I had a problem with was, “‘noun phrases’ is an odd typo for ‘sentences’. They’re not even close to each other on the keyboard.”)
I agree with this, and I think it’s already true. But I also think you worded it too softly to be in contradiction to my comment deletion (and more generally the implicit policy in my head). LW definitely does have said soft norm; I think allowing users to moderate their own posts, and users occasionally doing so, preserves that norm! Never deleting a comment, no matter where it lay on the dickish spectrum, would I think constitute a hard norm. The line I had a problem with was far beyond “not optimally packaged” and had nothing to do with using a “wrong shibboleths”.
I’ll note that I gave them plenty of time and opportunity to edit the comment; I requested it in the comments, I requested it in a PM, and I saw that they made other comments on the same post much later.
It spawned two comments (which you can still read, though I agree they’re harder to follow). I agree that the rest of said content was substantive and relevant. If there had actually been a whole lot more, I probably wouldn’t have deleted the parent. I just don’t think it was enough to tip the scale. And there’s tons of other similar discussion all over the comments, especially by Gwern. And like, everyone involved is free to reiterate the substantive and relevant content. (The option of copy-pasting someone’s comment feels weird to me, for reasons that I haven’t quite explicated and don’t feel super relevant.)
It really wasn’t that big of a comment. But also, I think our core disagreement might be here under “phrased a bit rudely”. I think the criterion I’m using implicitly here is that, if the entire purpose of the statement is to insult someone, then it’s out. That sentence did not have any other purpose. It’s not just phrasing that comes a bit too harsh. It was equivalent to, “You’re an idiot”. This is far from the side of the spectrum that requires finicky social norms. I’m not asking that people make sure that no one could be offended by what they’re saying. I’m not even saying that people shouldn’t say true, useful negative things about others. If it’s important to figuring out AI alignment that we discuss how smart a particular person is, then so be it. That’s a far cry from statements whose only purpose is to insult.
I’d actually be curious to know where the line is for you. If someone literally said, “You’re an idiot”, would you call that too dickish? Or what if someone’s comment had insulting profanities?
It might be worth to make sure that the author of a deleted comment can still read it so they can repost it on their shortform or a similar place.
Authors of deleted comments receive the text of the comment in a PM
Commenting to note that I agree (though I would put the matter in much stronger terms).
(These are all quantitative factors. If Gwern’s overall comment had sucked more, or his sentence had been way more egregious, I’d have objected a lot less to Alex’s call. But it does matter where we put rough quantitative thresholds.)