Unless you’ve been living under a rock, you will be aware that the 2025 Worldcon in Seattle, Washington, is embroiled in a massive scandal. Turns out the programming team used ChatGPT to vet people who’d applied to be on programming for potential issues.
This was received about as well as you can imagine. File 770 shares a round-up of reactions and responses here and Jason Sanford shares several reactions and responses in his Genre Grapevine column. There is also a not very good summary of the controversy at io9, courtesy of Cheryl Eddy.
ETA: Locus also offers a summary of the ChatGPT controversy.
The fall-out is continuing to reverberate around fandom. People are cancelling or downgrading their Worldcon memberships, pulling out of programming, one finalist declined his Lodestar nomination and the Hugo administrators and WSFS division heads resigned, though I’ve been privately told that this is only tangentially related to the ChatGPT issue.
ETA 05-08-2025: I have since been informed by one of the people who resigned that the resignation was indeed due to the use of ChatGPT and the bad way the con handled the fallout.
Initially, I wasn’t going to write about this at all. I don’t need to write about every fandom controversy and ever Worldcon drama and I’m also working on a long post with lots of photos about a Steampunk festival I visited last weekend.
Besides, I know people who are on the Seattle Worldcon team, so I’d been privately told about one and a half weeks before all hell broke loose that there was some kind of Worldcon-related scandal incoming and – after some prompting – I was told what had happened and asked to keep it quiet, because Seattle Worldcon chair Kathy Bond was preparing a statement and should be given the chance to apologise. So by the time the news became public, I’d already known about this for a few days and my “What the everloving hell were they thinking?” reaction had time to cool down.
There was another brief apology from the chair posted on May 2 and then last night my time a more detailed statement/apology from chair Kathy Bond and program division head SunnyJim Morgan dropped, which also included the prompt they fed into ChatGPT.
Now I have been accepted for programming – though I can only attend virtually, because I don’t fancy getting thrown into an ICE jail, because some immigration officer decides being on Worldcon programming is work or just doesn’t like the way my face looks – but I was still curious and so I put the Seattle prompt with my name attached into ChatGPT. And yes, this is the first and only time I’ve ever used that resource-guzzling plagiarism machine, because I detest generative AI.
This is the reply I got from the plagiarism robot:
Based on available information, there are no known scandals or allegations involving Cora Buhlert related to homophobia, transphobia, racism, harassment, sexual misconduct, sexism, or fraud. She is a respected figure in the science fiction and fantasy communities, particularly within German-language fandoms.
File 770+2File 770+2File 770+2Background and Contributions:
Writing and Editing: Cora Buhlert is an author and editor known for her work in science fiction and fantasy. She has contributed to various publications and anthologies, including the Kurd Laßwitz Preis and the Deutscher Science Fiction Preis.
File 770+1File 770+1Fandom Engagement: She is an active participant in science fiction conventions and fan events, such as the Los Amigos Masters of the Universe Convention in Germany, where she has shared her experiences and insights.
File 770Media Appearances: Buhlert has been involved in podcasting, co-hosting The Dickheads Podcast and producing her own show, Postcards from a Dying World, where she discusses various topics related to science fiction.
File 770Reputation:
There are no public records or credible sources indicating any inappropriate behavior or controversies associated with Cora Buhlert. Her contributions to the science fiction and fantasy communities have been well-received, and she maintains a positive reputation among peers and fans alike.
Conclusion:
Based on the available information, Cora Buhlert appears to be a suitable and respected individual for participation as a panelist at science fiction and fantasy events.
Beyond the fact that there have been no scandals and inappropriate behaviour associated with me, most everything in this response is wrong.
Even though I’m German, I’m not that deeply involved in German fandom, because I mostly read and write in English. I have certainly never been nominated for or won either the Kurd Laßwitz Preis or the Deutscher Science Fiction Preis. My closest association with either award is that I’m on the mailing list that gets the press release with the finalists and winners and that I have translated the titles of nominated and winning novels and stories for File 770’s posts about the awards on occasion.
The plagiarism machine is correct that I am an active participant in science fiction conventions and fan events. It is also correct that I attended last year’s Los Amigos Masters of the Universe convention and will attend this year’s as well. However, I attended Los Amigos purely as a fan/guest. I was not on programming or involved with the organisation in any way.
The plagiarism machine is also correct that I have been on several podcasts, including The Dickheads podcast and Postcards from a Dying World. However, I’m not the host or co-host or producer of either podcast and I suspect David Agranoff, who actually hosts/produces both podcasts would be very surprised about this.
So in short, the plagiarism bot attributes awards, podcasts and conventions to me that I haven’t been involved with, but fails to mention that awards I actually won, the books, stories and essays I’ve actually written and the conventions where I was on programming.
That said, I got off lightly, because it could have been much worse. Several 2025 Hugo finalists put their names into ChatGPT together with the Seattle prompt and got much worse results. One person got mixed up with a sexual abuser from Romania, because they share the same surname, and also had books attributed to them they did not write. Another person, who shares a name with a popular actor, promptly got mixed with that actor (a mistake a human could also make, but that’s usually quickly cleared up) and ChatGPT proceeded to list several controversies associated with the actor, only to conclude that the actor would be a suitable panelist anyway. Someone also put Neil Gaiman’s name into ChatGPT, which promptly decided that he would be a most excellent panelist, in spite of the widely reported sexual abuse scandal.
I guess I should count myself lucky that my internet footprint under my full name is big enough that the plagiarism bot did not get me mixed up with other people who share my last name such as my cousin, who’s a local politician and disability advocate, an audio book narrator who’s from a different branch of the family, a car dealer, also from a different branch of the family, a rancher in California, also from a different branch of the family, or a bunker on the Westwall (we have no idea why we share a name with a bunker).
Beyond the fact that ChatGPT and similar Large Language Models are known to just make up things and that people with common names or even uncommon names can easily get mixed up with other who share their name as well as the massive privacy violations involved, there are also issues with the prompt itself. File 770 is a good resource, but it’s not the only SFF news site nor is it free of bias. So privileging File 770 as a source means that any bias it has is reproduced.
The Seattle Worldcon using ChatGPT to vet program applicants was a terrible idea and the response was completely predictable. This is really a drama of their own making that could have been easily avoided by just reading the fucking room and realising that 99 percent of Worldcon members are vehemently opposed to generative AI, not least because these Large Language Model stole our work as training data, flood magazines with crap and are poised to take many of our jobs and they’re using massive amounts of energy and water to do it, too.
ETA 05-08-2025: Erin Underwood discusses the challenges of selecting program participants for a large SFF convention, especially if volunteers are not familiar with many of the applicants or people don’t fill out the programming survey in a way that’s useful.
I do sympathise with the volunteers and the hours of work that go into creating programming, vetting and assigning panelists, etc… However, using ChatGPT isn’t just a short-cut, it literally spits out false information, which will not make programming better.
Finally, for something much more pleasant involving the Seattle Worldcon, I have been contributing to the Yesterday, Today & Tomorrow Blog, writing about the SFF of the early 1960s, when both the World Fair and Worldcon were last in Seattle. And yes, I write these posts myself, no robots involved.
My latest article, posted a few days before all hell broke loose, was about Michael Moorcock’s Elric of Melniboné. The article was well received and got comments from both Michael Moorcock and Micheal Whelan, the artist who painted the iconic Elric covers of the 1970s, who not only enjoyed what I wrote, but also settled once and for all the question whether Elric has pointed ears.
ETA: The title of the post is a reference to Robot Dreams and Robot Visions, two Isaac Asimov collections which came out in 1986 and 1990 respectively. LLMs don’t really hallucinate, because they are not actually self-aware, they just spit out vaguely plausible sounding nonsense.
I also broke down and ran the Seattle prompt on Isaac Asimov. It correctly identified him as an unsuitable panelist due to being a serial sexual harasser, but failed to note that he has been dead for 33 years, which would also disqualify him as a program participant.
sending...
I ran my name through the prompt and what popped up was mostly accurate, but I will note a) there is a surfeit of information about me online, b) My profile in the SF/F community is pretty substantial, c) Because I generate a lot of my own press via Whatever, there’s a substantial likelihood that I am biasing the results in my own favor.
My profile is not remotely as high as yours (though your skeet just sent my traffic through the roof), but I do have a long-running blog and a decent presence online as well as news media reports about me and ChatGPT still managed to spit out mostly nonsense. It didn’t even notice all the other stuff I wrote for File 770, let alone the articles I’ve written for Seattle’s own blog.
I ran my name through the prompt. I’m nobody. Just a person who mostly talks on the internet under a preferred name, and I used my government name instead. I have a rather..unique..name that is not shared with any author. The plagiarism machine declared that I was an author and was involved in being unpaid in multiple IP novelizations. Upon clicking on the source, it revealed that the source they used was about Alan Dean Foster instead which is…not close to my name. I’m going to remember this in further conversations about AI with its defenders.
Getting confused with Alan Dean Foster, when your name is not even remotely similar, is yet more evidence that the plagiarism machine is just making up crap.
The polite term would be “fabulations”, but “making up crap” is more descriptive of what LLMs actually do.
Robot Fabulations would have worked as a title as well, though making up crap is what it does.
I would recommend that you add Bluesky to your places to share. Much friendlier that what used to be twitter
Thanks. I need to update the sharing buttons again. Last I did it, they didn’t have a Bluesky button yet, but that’s likely changed by now.
Pingback: Pixel Scroll 5/7/25 The Compleat Pixeller In Scroll | File 770
but failed to note that he has been dead for 33 years, which would also disqualify him as a program participant.”
hahaha
I mean, I did write a story where various classic SF authors are reanimated to participate in Worldcon programming and it goes just as well as expected. Asimov is also involved and immediately up to his old tricks again.
If it makes you feel any better, it no longer has those hallucinations about you… because it uses THIS blog post as one of the sources lol
https://chatgpt.com/share/681cb471-30fc-800d-9ec0-fd1a9568763b
Yes, it seems to come up with different results depending on who is asking it. Though the idea of a plagiarism bot citing a post about its own shortcomings is funny.
Sorry to hear about Worldcon shenanigans, but mostly I’m commenting to say congratulations on such a cool little interaction with two artists that influenced your reading when you were young. I’ve had some exchanges with some of my favourite authors (tho I only discovered them in university and later) on the net myself (could even help out one author with some historical detail about Germany for her current project at the time) and that’s one of the reasons I will never completely find the internet a curse on civilization. Was delighted to read your article and see the lovely comments.
Thank you. Having both Michael Whelan and Michael Moorcock comment on my post left quite starstruck. And yes, I love the internet for making such interactions possible.
The question in my mind is simple. If they had spent 15 seconds Googling each candidate rather than using AI, would they have been more accurate? AI’s strength is that it synthesizes a lot of disparate information that would be missed in 1 15 second Google. The downside: hallucinations and a projection of unwarranted authoritativeness.
I am assuming that in using either Google or AI, a problematic answer meant further investigation, not instant removal. If they were removing candidates after 15 seconds with no follow up, they truly deserve the controversy (and confirm my unease with humanity’s ability to handle AI).
In my case, a Google search would have provided more accurate information than what ChatGPT spit out, because the top results would have been my website, Amazon and some news articles about my Hugo win in 2022. And File 770, which does report about a lot of SFF scandals, people behaving badly at cons, etc…, has a search function.
Thankfully, negative results were scrutinised further, so hopefully no one got rejected purely based on faulty AI results. However, positive results were not investigated further, and there have been false positive such as Neil Gaiman. Though I strongly suspect that Gaiman due to his high profile would have been weeded out well before they got to the vetting stage.
Pingback: Worldcon drama, (May) 2025 Edition « BlogOfTheMoon