The part of me that used to work for a survey research nonprofit is screaming into a pillow
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
"we don't need to get actual data to get an accurate average. we'll just average up all the already averaged data. it will be just as good."
narrator: "someone flunked their stats 101 course..."
-
R relay@relay.infosec.exchange shared this topic
-
@foobarsoft @hannah I bet they are hip deep in "digital twin" kool-aid.
@andygates @foobarsoft bingo, those are mentioned later in the article
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah this truly is the death of all information integrity everywhere.
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah Aaaaaaah.
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah Feeding the dog it's own vomit. Force-feeding.
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah qualtrics floated this (and maybe went through with it idk) right after chatgpt first went live
What hilarious is that this agentic things is plausibly *worse* since you're having simulations fill out the survey instead instead of using an llm to generate a bazillion responses. The responses will still be just as bad, but at a huge markup of energy costs
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
nothing is real
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah Wow, sock-puppeting at scale.
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
-
10/10 LLMs think that this is a tremendous idea.
I have a hard time getting over this.
I tried irony. Sarcasm. I duckduckgoed the word.
Maybe the article misunderstood something? No, doesn't look that way.
Sure, there is bias in any sort of polling or study. The way you phrase things can influence how they are answered.
But this is just a dice roll with more words.
No wait, dice rolls can't be manipulated that easily to skew opinion.
-
@cinebox marc andreesen just venmoed you $50m
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah "silicon sampling" inventing fancy new words for synthetic data
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah Perhaps I'm just missing something obvious here, but...isn't the whole point of conducting surveys to gather data from people? If the only entities being surveyed are A.I. agents that mimic human speech and which could be programmed to say anything, how is it meaningful?
It seems as meaningful to me as "Our survey data says 95% of people approve our product. We surveyed 1000 people whom we paid $100 each to say positive things about us."
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah next up: instead of spending budget money on organizing elections, we'll just run 400 millions of AI agents, give them personas representing the US demographic, and ask them who would they vote for.
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah "tantalizing" ain't the word.
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah Sounds like someone read about the bootstrap without actually understanding why it works.
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah@posts.rat.pictures what the actual fucking fuck? So you "simulate" a poll and it generates whatever bullshit outcome you ask for
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah That's a lot of words ot say: we gonna make shit up
-
The part of me that used to work for a survey research nonprofit is screaming into a pillow
@hannah "Four out of five LLMs recommend Trident sugarless gum."
-
@hannah Wait, I've got a better idea that will still cut their polling costs in half: we survey people but with half the standard sample size, then we run it all through a photo copier and tabulate the results. We'll get the normal sample size and accuracy!
Also, copy machines use a lot less power than AI, its win-win!
