I don’t get it. Instead of sending me the chatbot output, just send me your goddamn prompt. Cut out the middleman.
22 thoughts on “People keep sending me AI slop that they want me to post on the blog.”
I like the term “meat proxy” from this article: https://gruhn.me/blog/2026-08-03/. Hopefully a strong social taboo will develop around this type of behavior.
No, what people are doing to me is much worse. They’re sending me AI slop, but not labeling it as such. So I first have to waste the time to read it and figure out that it is slop.
I guess there might be a way to pass my email through an AI-detection filter and then automatically flag anything containing chatbot output? I don’t know how to set up such a filter; indeed, I already have problems with Columbia’s email spam filter, which doesn’t allow me to put in a blacklist or a whitelist, and as a result I keep getting some of the same spam every day.
Andrew, I don’t think they have email integration yet, but if you make yourself a Pangram account (https://www.pangram.com/) you can then just go there, paste in the text, and see if it returns mostly AI-generated before you waste time reading it. You get a certain number of free credits per day which I suspect is enough for what you need.
I didn’t realize it was unlabeled – that’s definitely worse. Ideally we’d have spam filters for AI at this point, but Microsoft has been more focused on producing the slop than dealing with its consequences.
Jessica’s comment about Pangram had me thinking about setting up an automated system to scan and label emails, but I think it would become prohibitively expensive ($100/mo) depending on how many emails you receive.
Yeah, Microsoft has been annoyingly AI-forward in ways that don’t make a lot of sense to me. But then a lot of enterprise software companies are.
Coding is one thing, but why is Adobe constantly offering to summarize or simplify my work documents for me? I need to be reading these things for details, not relying on some probabilistic summary.
What puzzles me is, that given the nature of Andrew’s work, one should think that he would be a rather unlikely target for this [expletive deleted]. Are the email addresses of the senders for real?
(one note of caution, though: emails from the Free University of Berlin are labelled ‘fu-berlin’; we’re almost in Odd Couple territory here.)
I send Andrew emails every now and then — this blog itself is a good advertisement platform for your work (of course if it makes sense for Andrew to share on the blog). No different than other types of SEO.
this is how the world is these days. it’s a big issue at work. I’ve been thinking about this in the context of the arday scandal – of course, these days, he wouldn’t plagiarize, he’d just get a chatbot to output “his” paper instead.
what bothers me is that it seems like no one really cares. our society has strong – probably historically unusual – norms against plagiarism. the developing norms against ai generated text seem much weaker. so we’re drifting towards a world in which whatever moral flaws people feel like they’re filtering out when arday / agus resign are no longer detectable.
why is it suddenly fine to pass off something else’s thoughts as your own?
I think the worst thing that Arday did, from a scholarly perspective, was to fabricate survey responses. He did it by plagiarism, but I think the big problem was the fabrication.
Sociologists Alice Goffman and Sudhir Venkatesh have been credibly accused of fabricating their research, making things up and reporting them in their books as if they had really occurred. It seems they did this using their imagination. But if they’d done it using plagiarism (or, with a modern twist, using a chatbot), it would also be scholarly misconduct.
So to me the appropriate analogy to using a chatbot to fake research is not plagiarism but rather “curbstoning” (that’s what it’s called when survey interviewers come to an empty house and, instead of coming back later to interview someone who’s at home, they sit on the curbstone and fill out the survey form themselves).
I’m not so worried about people’s passing off someone else’s thoughts as their own; I’m more worried about them reporting things that never happened, or quoting people as saying things they never said.
Back in the old (and I mean very old) days (circa 1980) when I lived in Boulder, I remember friends who made extra money working for a market research firm doing surveys. They had a quote of demographics they needed: e.g. 35 year old single white women. And survey respondents would earn a small fee. So, they would approach someone (say a 65 year old Hispanic male) and say “you can earn $5 and you are a 35 year olds single white woman, right?” It was easier than trying to find the requested demographic.
Today it is so much easier! No need to even use post-survey weighting, just have AI produce the desired responses. Efficiency gain!
I propose an annual Dolezal Prize, presented by Ali G.
And then there would be the SWAG award (Stapel,Wansink,Ariely,Guéguen) presented alternately by Malcolm Gladwell and Jordan Peterson.
Gimme Stuart Hall anytime, but spare me …
“Today it is so much easier! No need to even use post-survey weighting, just have AI produce the desired responses. Efficiency gain!”
Huh??? Dale’s sounding like me??? Have times changed???
Seriously, though; people here understand that LLM based AI is based on statistical random text generation (albeit nowadays with external looping through some sorts of checking, e.g. Lean), so that there are certain expected problems with it.
But the vast majority of people in the real world don’t understand this, and now think that “AI” is “real”. So, for example, Noah Smith (whom I see as having previously been largely reasonable, but as having recently started going off the rails*) sees the anti-AI (especially the anti-AI-slop folks) as unreasonable anti-tech troglodytes. Which we’re not; we’re technically correct: an edifice built on statistical random text generation is seriously problematic.
Sigh.
By the way, somewhat on-topic, here’s a linguist who has some ideas on how to detect AI slop. (I’m a fan of this bloke: consider thynself warned.)
*: He lived in Japan for a few years and is a Japan fan. But he doesn’t undestand that while the Japanese stock market is doing great, many large Japanese are doing great, and the _average_ income is going up, the majority of the Japanese population is scrod, since the _MEDIAN INCOME_ has been falling for the last 30 years. (Also, the failure rate of small and medium-size companies is the highest it has ever been.)
David
We have no disagreement about the dangers of AI nor about the likely misuse by a high percentage of people. Nor about the benefits of that misuse accruing to the wealth of the purveyors of hype. What we do disagree with is the usefulness of AI when properly used. I take your position to be that there is almost no such thing as “properly used.” My own position is that it is more useful than I had imagined. I can show you dozens of cases (I am in the process of writing a book, so I have many such examples) where AI has allowed me to accomplish things that I could not do on my own – things of value. My own personal experience with AI does not modify my pessimistic view of how it is likely to be (mis)used, but I think it leads me to different personal conclusions about the usefulness of AI than you express.
Curbstoning sounds deplorable, but not as bad as curbstomping.
The term, “curbstoning” is so evocative that it also has another meaning different from what was just posted:
“survey interviewers come to an empty house and, instead of coming back later to interview someone who’s at home, they sit on the curbstone and fill out the survey form themselves”
——————————————————————————————————-
“1a. Curbstoning is when a dealer poses as a private seller to sell a car. By curbstoning, an unethical dealer can avoid having to comply with the regulations that apply to dealers. To a buyer, this could mean buying a car that has a salvaged title (a car that’s been declared a total loss by an insurance company).
1b. In British English, “kerbstoning” has a literal meaning completely absent from American English. In Formula 1 and European motorsport, “kerbstoning” refers to a driver aggressively running over the raised, painted curbs (kerbs) at the edge of a racetrack corner to find a faster racing line.”
Yeah, that was before I knew to check for this sort of thing. Now I’m more aware.
I don’t think that everyone who sends me slop is spamming me or has bad intentions. There’s a lot of that, but I assume that there are also people with legitimate questions who just consider it a norm to write their emails using the chatbot. They might be feeding it prompts, or maybe they’re in a text processing environment where at any point they can click a button to have the chatbot complete the paragraph. Or maybe it’s something else. In any case, I hate it, because it produces out-of-focus slop that looks like something interesting and maybe important that’s just not quite comprehensible.
tl;dr. I’d like the spammers to stop entirely. I’d like the sincere people to just send me their prompts; the effort to make things clean and readable using the chatbot actually just makes things worse. The chatbot might well be useful for giving clean instructions on how to fix your toaster or how to make a certain graph in R, but it’s not working on tough questions where you need expert help.
a) What folks perceive as ‘generative’ AI – oh aye! – is a combination of the Dada engine and a polished ELIZA frontend. (NB Tried it several times, found it wanting.)
b) The only LLMs I find worthy of study are children from the last trimester in the womb to their first four post-partum years – Little Learning Machines.
Here is my prompt: Write a short invitation text for a statistic professor at Columbia University for the Journal of Albanian Conspiracy Theories on the topic of statistics as a conspiracy (either on its mathematical principles or in its application in history, politics, sociological research, lately information technology and Al). The article can be a prompt, AI slop, plagiarized or montage. The author can choose his own nome de guerre.
I like the term “meat proxy” from this article: https://gruhn.me/blog/2026-08-03/. Hopefully a strong social taboo will develop around this type of behavior.
Tim:
No, what people are doing to me is much worse. They’re sending me AI slop, but not labeling it as such. So I first have to waste the time to read it and figure out that it is slop.
I guess there might be a way to pass my email through an AI-detection filter and then automatically flag anything containing chatbot output? I don’t know how to set up such a filter; indeed, I already have problems with Columbia’s email spam filter, which doesn’t allow me to put in a blacklist or a whitelist, and as a result I keep getting some of the same spam every day.
Andrew, I don’t think they have email integration yet, but if you make yourself a Pangram account (https://www.pangram.com/) you can then just go there, paste in the text, and see if it returns mostly AI-generated before you waste time reading it. You get a certain number of free credits per day which I suspect is enough for what you need.
Andrew:
I didn’t realize it was unlabeled – that’s definitely worse. Ideally we’d have spam filters for AI at this point, but Microsoft has been more focused on producing the slop than dealing with its consequences.
Jessica’s comment about Pangram had me thinking about setting up an automated system to scan and label emails, but I think it would become prohibitively expensive ($100/mo) depending on how many emails you receive.
Yeah, Microsoft has been annoyingly AI-forward in ways that don’t make a lot of sense to me. But then a lot of enterprise software companies are.
Coding is one thing, but why is Adobe constantly offering to summarize or simplify my work documents for me? I need to be reading these things for details, not relying on some probabilistic summary.
https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content
Anthropic is going to start subtlety watermarking their bot’s text.
What puzzles me is, that given the nature of Andrew’s work, one should think that he would be a rather unlikely target for this [expletive deleted]. Are the email addresses of the senders for real?
(one note of caution, though: emails from the Free University of Berlin are labelled ‘fu-berlin’; we’re almost in Odd Couple territory here.)
I send Andrew emails every now and then — this blog itself is a good advertisement platform for your work (of course if it makes sense for Andrew to share on the blog). No different than other types of SEO.
this is how the world is these days. it’s a big issue at work. I’ve been thinking about this in the context of the arday scandal – of course, these days, he wouldn’t plagiarize, he’d just get a chatbot to output “his” paper instead.
what bothers me is that it seems like no one really cares. our society has strong – probably historically unusual – norms against plagiarism. the developing norms against ai generated text seem much weaker. so we’re drifting towards a world in which whatever moral flaws people feel like they’re filtering out when arday / agus resign are no longer detectable.
why is it suddenly fine to pass off something else’s thoughts as your own?
Jcp:
I think the worst thing that Arday did, from a scholarly perspective, was to fabricate survey responses. He did it by plagiarism, but I think the big problem was the fabrication.
Sociologists Alice Goffman and Sudhir Venkatesh have been credibly accused of fabricating their research, making things up and reporting them in their books as if they had really occurred. It seems they did this using their imagination. But if they’d done it using plagiarism (or, with a modern twist, using a chatbot), it would also be scholarly misconduct.
So to me the appropriate analogy to using a chatbot to fake research is not plagiarism but rather “curbstoning” (that’s what it’s called when survey interviewers come to an empty house and, instead of coming back later to interview someone who’s at home, they sit on the curbstone and fill out the survey form themselves).
I’m not so worried about people’s passing off someone else’s thoughts as their own; I’m more worried about them reporting things that never happened, or quoting people as saying things they never said.
Back in the old (and I mean very old) days (circa 1980) when I lived in Boulder, I remember friends who made extra money working for a market research firm doing surveys. They had a quote of demographics they needed: e.g. 35 year old single white women. And survey respondents would earn a small fee. So, they would approach someone (say a 65 year old Hispanic male) and say “you can earn $5 and you are a 35 year olds single white woman, right?” It was easier than trying to find the requested demographic.
Today it is so much easier! No need to even use post-survey weighting, just have AI produce the desired responses. Efficiency gain!
I propose an annual Dolezal Prize, presented by Ali G.
And then there would be the SWAG award (Stapel,Wansink,Ariely,Guéguen) presented alternately by Malcolm Gladwell and Jordan Peterson.
Gimme Stuart Hall anytime, but spare me …
“Today it is so much easier! No need to even use post-survey weighting, just have AI produce the desired responses. Efficiency gain!”
Huh??? Dale’s sounding like me??? Have times changed???
Seriously, though; people here understand that LLM based AI is based on statistical random text generation (albeit nowadays with external looping through some sorts of checking, e.g. Lean), so that there are certain expected problems with it.
But the vast majority of people in the real world don’t understand this, and now think that “AI” is “real”. So, for example, Noah Smith (whom I see as having previously been largely reasonable, but as having recently started going off the rails*) sees the anti-AI (especially the anti-AI-slop folks) as unreasonable anti-tech troglodytes. Which we’re not; we’re technically correct: an edifice built on statistical random text generation is seriously problematic.
Sigh.
By the way, somewhat on-topic, here’s a linguist who has some ideas on how to detect AI slop. (I’m a fan of this bloke: consider thynself warned.)
https://www.youtube.com/watch?v=ORgKY9AlybA
*: He lived in Japan for a few years and is a Japan fan. But he doesn’t undestand that while the Japanese stock market is doing great, many large Japanese are doing great, and the _average_ income is going up, the majority of the Japanese population is scrod, since the _MEDIAN INCOME_ has been falling for the last 30 years. (Also, the failure rate of small and medium-size companies is the highest it has ever been.)
David
We have no disagreement about the dangers of AI nor about the likely misuse by a high percentage of people. Nor about the benefits of that misuse accruing to the wealth of the purveyors of hype. What we do disagree with is the usefulness of AI when properly used. I take your position to be that there is almost no such thing as “properly used.” My own position is that it is more useful than I had imagined. I can show you dozens of cases (I am in the process of writing a book, so I have many such examples) where AI has allowed me to accomplish things that I could not do on my own – things of value. My own personal experience with AI does not modify my pessimistic view of how it is likely to be (mis)used, but I think it leads me to different personal conclusions about the usefulness of AI than you express.
Curbstoning sounds deplorable, but not as bad as curbstomping.
The term, “curbstoning” is so evocative that it also has another meaning different from what was just posted:
“survey interviewers come to an empty house and, instead of coming back later to interview someone who’s at home, they sit on the curbstone and fill out the survey form themselves”
——————————————————————————————————-
“1a. Curbstoning is when a dealer poses as a private seller to sell a car. By curbstoning, an unethical dealer can avoid having to comply with the regulations that apply to dealers. To a buyer, this could mean buying a car that has a salvaged title (a car that’s been declared a total loss by an insurance company).
1b. In British English, “kerbstoning” has a literal meaning completely absent from American English. In Formula 1 and European motorsport, “kerbstoning” refers to a driver aggressively running over the raised, painted curbs (kerbs) at the edge of a racetrack corner to find a faster racing line.”
not to be confused with “kerb crawling” ;-)
I assume Andrew already saw this, but Alex’s text from this post was all AI-generated: https://statmodeling.stat.columbia.edu/2026/07/29/over-coverage-caught-by-pre-registration-47-of-56-inside-a-stated-50-interval/
Ai,
Yeah, that was before I knew to check for this sort of thing. Now I’m more aware.
I don’t think that everyone who sends me slop is spamming me or has bad intentions. There’s a lot of that, but I assume that there are also people with legitimate questions who just consider it a norm to write their emails using the chatbot. They might be feeding it prompts, or maybe they’re in a text processing environment where at any point they can click a button to have the chatbot complete the paragraph. Or maybe it’s something else. In any case, I hate it, because it produces out-of-focus slop that looks like something interesting and maybe important that’s just not quite comprehensible.
tl;dr. I’d like the spammers to stop entirely. I’d like the sincere people to just send me their prompts; the effort to make things clean and readable using the chatbot actually just makes things worse. The chatbot might well be useful for giving clean instructions on how to fix your toaster or how to make a certain graph in R, but it’s not working on tough questions where you need expert help.
a) What folks perceive as ‘generative’ AI – oh aye! – is a combination of the Dada engine and a polished ELIZA frontend. (NB Tried it several times, found it wanting.)
b) The only LLMs I find worthy of study are children from the last trimester in the womb to their first four post-partum years – Little Learning Machines.
Max Read recently expressed a similar sentiment, although his position is less utilitarian (and less laconic!) than yours:
https://www.patreon.com/MaxRead/posts/just-send-me-167205259
Here is my prompt: Write a short invitation text for a statistic professor at Columbia University for the Journal of Albanian Conspiracy Theories on the topic of statistics as a conspiracy (either on its mathematical principles or in its application in history, politics, sociological research, lately information technology and Al). The article can be a prompt, AI slop, plagiarized or montage. The author can choose his own nome de guerre.