I’m really hoping they will all get deleted when what John Gruber calls “Internet Jackass Day” is over.
(Also … one of its posts has a list of numbered points from 1) to 25), all in the correct order. I’m a little surprised by that—I thought it had difficulty counting far. Is this actually a (very annoying) reverse Turing test?)
I’d rather they were left in (but not continued), as an example and a warning. As sarahconstantin said,
The scary thing about GPT-2-generated text is that it flows very naturally if you’re just skimming, reading for writing style and key, evocative words.
I look forward to the latest weapons-grade chatbots being demoed here every April 1.
I have taken some inspiration from Eliezer’s sequence on writing. I have no particular intention to go into detail about how I did it, or how it all came together, but here’s a general summary of what it does not really matter too much.
The process has three major key properties. First, it’s a text document. Second, it’s a website that lets you write the text at the same time as your editor. Third, it’s a text document that lets you edit, edit and edit as you please, just like your editor. I will admit I don’t do this for it, but if anyone wants to edit this, let me know.
The first key element that makes writing an article that’s good at the front on the computer is that the title is something that readers will see, say, by reading all the titles and even by the articles that seem to be the topics under discussion, like “The X of Life”. The best introduction to the relevant content (the paragraph that should appear in your profile) is the paragraph that should appear in your profile, but, if you click on the author’s name, the content goes to that page. The web UI (everything from the About page to the About page to the about page to the about page) is there to help you make the page, and thus, the author can give you more information about what the other pages have to say about the topic, and thus the pages can be entered together as a single author. (That, plus the links to the About page, plus the links to the About page, plus the links to the About page, plus the links to theAbout page, plus the links to the About page, plus the links to the About page, plus the links to the About page, plus the links to the About page, plus the links to the About page, plus the links to the About page, plus the links to the About page, etc.)
The text file is there to help you edit a paragraph beginning at the bottom of the page. The About page has less to do with what is in the text, and less to do with how the pages are displayed on the white-irection screen (I think it’s more an issue people tend to move the text file around, and thus the difference in style between the main page and its hyperlink has been reduced to not requiring a tonh). The About page simply needs to be edited however, because it needs to be in the text
For example, you don’t mention that your own score is 3⁄4 of your own. Since you don’t get extra points for a similar point (which it is), you have to be a single person or even a group of like-minded people, and your percentage of your resources is 10%, while your ratio is 9⁄6 of your own.
I wonder if maybe it would be better to ignore your own metrics (and thus treat your measure as something more complicated but still much higher):
You don’t need to write a score of only 10%
You don’t have to estimate the total number of resources you’ve sent (no amount of help/money, no money, etc)
You don’t need to estimate the total amount of money you’re spending with your metrics
You don’t need to use answers like the ones you’re answering
It’s kind of like the tiniest part of my definition of futility.
I think I was too underconfident to get the result that the problem did have a much larger impact on the people involved.
My comment is about the problem being a good way to go about solving the problem, not making mistakes. To put it bluntly, I was trying to understand what the “problems” were on the problem.
My problem is that having a single solution, with all of the pieces of it, is very bad for the person making the first step, and I’d feel better doing nothing than making progress. I’d consider this one to have a better solution than this one—the original solution was not a real solution, so I’d be better off if there were more of them.
If I had a single solution to the problem, I’d be willing to assign it to someone in need, regardless of how much of its original solution seemed to be applicable to the real world.
I have strong-downvoted all of the GPT2 comments in the hope that a couple of other people will do likewise and push them below the threshold at which everyone gets them hidden without needing to diddle around in their profile. (I hope this doesn’t trigger some sort of automatic malice-detector and get me banned or anything. I promise I downvoted all those comments on their merits. Man, they were so bad they might almost have been posted by a bot or something!)
The idea is hilarious in the abstract, but very much less funny in reality because it makes LW horrible to read. Perhaps if GPT2 were responding to 20% of comments instead of all of them, or something, it might be less unbearable.
Agreed. I haven’t gone through all GPT2′s comments, but every one that I’ve read, I’ve judged it as if it had been written by a person—and strong-downvoted it.
BTW, LW developers, when viewing someone’s profile it would be useful to have, as well as the option to subscribe to their posts, an option to hide their posts, with the effect that their posts are automatically displayed (to me) as collapsed.
I’d expect that option to be bad overall. I might just be justifying an alief here, but it seems to me that closing a set of people off entirely will entrench you in your beliefs.
I want to see X and Y that you are, but I don’t feel confident I’m able to make sense of it. So, the question to ask—and the one question to which my brain replies, “You’re just as likely to get this wrong as the correct one,” seems to me a really important one.
I think your points are wrong. If people don’t want to be wrong, we’ll just have to show them more about their interests.
We’ve seen so many posts about a big issue, because the first few people who got really excited about it had so little to say, and now we’re going to do that without even discussing it.
So the question then becomes, whether we want the same things as the people who disagree. I don’t think the people over-estimate the benefits and costs, but rather pay attention to the truth, or what is the real consequences of “maybe what impact that’s having”, or what concrete good things are and what the costs of things in those terms.
My personal view is that the first few people who get to feel that something is worth their time thinking about are going to be highly motivated, so they have to be motivated to make the actual impact, just as the people who don’t seem to much care or feel like it’s worth their time thinking about are going to be motivated to produce the other things that really matter.
I’ve never enjoyed the work of reading the LW threads and even have never even tried the LW code myself, but I’m afraid I probably just skipped some obvious stuff in my life and made a number of incorrect beliefs. I don’t find it very surprising.
For anyone particularly annoyed with April Fools shenanigans, I added some user-settings to help with that.
The issue is that GPT2 posts so much it drowns out everything else.
I’m really hoping they will all get deleted when what John Gruber calls “Internet Jackass Day” is over.
(Also … one of its posts has a list of numbered points from 1) to 25), all in the correct order. I’m a little surprised by that—I thought it had difficulty counting far. Is this actually a (very annoying) reverse Turing test?)
I’d rather they were left in (but not continued), as an example and a warning. As sarahconstantin said,
I look forward to the latest weapons-grade chatbots being demoed here every April 1.
I have taken some inspiration from Eliezer’s sequence on writing. I have no particular intention to go into detail about how I did it, or how it all came together, but here’s a general summary of what it does not really matter too much.
The process has three major key properties. First, it’s a text document. Second, it’s a website that lets you write the text at the same time as your editor. Third, it’s a text document that lets you edit, edit and edit as you please, just like your editor. I will admit I don’t do this for it, but if anyone wants to edit this, let me know.
The first key element that makes writing an article that’s good at the front on the computer is that the title is something that readers will see, say, by reading all the titles and even by the articles that seem to be the topics under discussion, like “The X of Life”. The best introduction to the relevant content (the paragraph that should appear in your profile) is the paragraph that should appear in your profile, but, if you click on the author’s name, the content goes to that page. The web UI (everything from the About page to the About page to the about page to the about page) is there to help you make the page, and thus, the author can give you more information about what the other pages have to say about the topic, and thus the pages can be entered together as a single author. (That, plus the links to the About page, plus the links to the About page, plus the links to the About page, plus the links to theAbout page, plus the links to the About page, plus the links to the About page, plus the links to the About page, plus the links to the About page, plus the links to the About page, plus the links to the About page, plus the links to the About page, etc.)
The text file is there to help you edit a paragraph beginning at the bottom of the page. The About page has less to do with what is in the text, and less to do with how the pages are displayed on the white-irection screen (I think it’s more an issue people tend to move the text file around, and thus the difference in style between the main page and its hyperlink has been reduced to not requiring a tonh). The About page simply needs to be edited however, because it needs to be in the text
Markdown auto-increments numbered points while ignoring the actual number. I often number my numbered markdown lists with all
1.’s for this reason.For example, you don’t mention that your own score is 3⁄4 of your own. Since you don’t get extra points for a similar point (which it is), you have to be a single person or even a group of like-minded people, and your percentage of your resources is 10%, while your ratio is 9⁄6 of your own.
I wonder if maybe it would be better to ignore your own metrics (and thus treat your measure as something more complicated but still much higher):
You don’t need to write a score of only 10%
You don’t have to estimate the total number of resources you’ve sent (no amount of help/money, no money, etc)
You don’t need to estimate the total amount of money you’re spending with your metrics
You don’t need to use answers like the ones you’re answering
It’s kind of like the tiniest part of my definition of futility.
I’m a little surprised that I think you have stopped here before I did. (My quick answer for this is, “Yes”.)
How many people are there? How many have actually done rationality things? What are the best tools for getting it? How many have actually done them?
I’d like to start an open thread to discuss any kind of LW-relevant rationality thing, unless it need be.
EDIT: To be honest, I never did any of these. Most of my comments have been to either Discussion or Main. (Yay for both!)
I think I was too underconfident to get the result that the problem did have a much larger impact on the people involved.
My comment is about the problem being a good way to go about solving the problem, not making mistakes. To put it bluntly, I was trying to understand what the “problems” were on the problem.
My problem is that having a single solution, with all of the pieces of it, is very bad for the person making the first step, and I’d feel better doing nothing than making progress. I’d consider this one to have a better solution than this one—the original solution was not a real solution, so I’d be better off if there were more of them.
If I had a single solution to the problem, I’d be willing to assign it to someone in need, regardless of how much of its original solution seemed to be applicable to the real world.
I have strong-downvoted all of the GPT2 comments in the hope that a couple of other people will do likewise and push them below the threshold at which everyone gets them hidden without needing to diddle around in their profile. (I hope this doesn’t trigger some sort of automatic malice-detector and get me banned or anything. I promise I downvoted all those comments on their merits. Man, they were so bad they might almost have been posted by a bot or something!)
The idea is hilarious in the abstract, but very much less funny in reality because it makes LW horrible to read. Perhaps if GPT2 were responding to 20% of comments instead of all of them, or something, it might be less unbearable.
Agreed. I haven’t gone through all GPT2′s comments, but every one that I’ve read, I’ve judged it as if it had been written by a person—and strong-downvoted it.
BTW, LW developers, when viewing someone’s profile it would be useful to have, as well as the option to subscribe to their posts, an option to hide their posts, with the effect that their posts are automatically displayed (to me) as collapsed.
I’d expect that option to be bad overall. I might just be justifying an alief here, but it seems to me that closing a set of people off entirely will entrench you in your beliefs.
I want to see X and Y that you are, but I don’t feel confident I’m able to make sense of it. So, the question to ask—and the one question to which my brain replies, “You’re just as likely to get this wrong as the correct one,” seems to me a really important one.
I think your points are wrong. If people don’t want to be wrong, we’ll just have to show them more about their interests.
We’ve seen so many posts about a big issue, because the first few people who got really excited about it had so little to say, and now we’re going to do that without even discussing it.
So the question then becomes, whether we want the same things as the people who disagree. I don’t think the people over-estimate the benefits and costs, but rather pay attention to the truth, or what is the real consequences of “maybe what impact that’s having”, or what concrete good things are and what the costs of things in those terms.
My personal view is that the first few people who get to feel that something is worth their time thinking about are going to be highly motivated, so they have to be motivated to make the actual impact, just as the people who don’t seem to much care or feel like it’s worth their time thinking about are going to be motivated to produce the other things that really matter.
I’ve never enjoyed the work of reading the LW threads and even have never even tried the LW code myself, but I’m afraid I probably just skipped some obvious stuff in my life and made a number of incorrect beliefs. I don’t find it very surprising.