for the record im not technially 100% anti-AI, in the sense that its a broad category of tech being lumped under one umbrella term so it feels over-zealous to say i hate all of it all the time forever. but i also think trying to discuss what it actually IS good for is difficult right now when i cant take one step without something trying to convince me to use chatgpt to summarize my life and speed up my hobbies and turn my friends into chatbots and optimize my life into oblivion. i am certain there is nuance to the topic but can we stop cramming the square peg into the round hole before you start trying to sell me on the legitimate benefits of the square peg. please.
Neural Nets have existed for decades and are genuinely useful. It's a form of AI that recognizes patterns, and can do stuff like identify cancer cells, tell whether an egg is fertilized or not, detect fraud, and optimize routes.
Those are Expert Systems, tuned to do exactly one thing. If you (say) ask a medical expert system a question about financial law, it's useless. The autopilot that flies a 787 has no idea how to drive a truck on the freeway. A Coulter Counter is excellent at identifying lymphocytes in a blood sample but can't predict the next card in a blackjack game.
And so on.
The problem with so-called generalized AI (AGI) is that we don't have that yet. It doesn't exist. It MIGHT some day, but AGI has been "10 years away" since the 1980s. The goals keep moving as we learn more about how people and machines process data.
But the current crop of AI techbros have been selling generative Large Language Model AI (LLM) as AGI because generative systems do a good job of faking it. There's no actual thought going on, merely the illusion of thought via predicting the next word in a sentence accurately.
If you let a human toddler listen to 800 hours of YouTube car influencer videos, that toddler might end up sounding like a car influencer. They'd parrot horsepower numbers and 0 to 60 times, mention EV range and MSRP numbers.
But they wouldn't understand any of it.
That's ChatGPT.
And yeah, it's worse than useless because it doesn't even know when it's lying or hallucinating. It just babbles convincingly until you stop it.
But for techbros to make money selling that as "AI"? It's the perfect scam, especially if you don't understand how it works.
I was among the participants that were going to exhibit my comic at Shortbox Comic Fair 2025. I applied to this fair due to its reputation and very few, almost no bad press about it. The only example I could find is the organizer's anti-competition statement here. Only during the course of participating in this fair, I learned why the Shortbox brand was so clean: the sole organizer, Zainab, would often send out emails to participants, scolding them and telling them to NOT say anything bad about their own Shortbox comics, or say anything against her decision throughout the fair. She was dictating our tone in both public and private spaces.
As for my case, I only communicated with Zainab via email. I tried to be as transparent with my progress as possible, though I was often met with a dismissive tone. One month prior to the deadline, I asked Zainab if my comic could have an 'extended' version, which I can link to Shortbox' customers after the fair so they can access it for free. She didn't comment or review my comic, but copy-pasted a statement from the participant guidebook about "exhibiting work post-Shortbox" (to summarize, we're not allowed to lower our prices or put up the comic for free immediately after the fair ends, an example can be seen here). Instead of a vendor fee, 2025 SBCF participants would have to deal with a 'drop out' fee, which applied to me later.
The five months after the Lunar New Year in March 2025, the comic for Shortbox was my entire priority. I took no breaks to finish the comic as fast as I can. During this period I didn’t have any income from commissions or freelance, I wanted to put my all into the comic even if the sales in October were close to nothing. All the risks were on the artists who couldn’t get any pay for the comic making period, while the organizer had little risk on it if there were only a few of us who couldn't make it to October.
I submitted my comic in 65 pages, all proofread and sent in at the end of July which was the deadline. My comic previews were even posted on Shortbox page. In mid August, Zainab got back to me to tell me that my comic is ineligible for submission because it's 'incomplete', and she demanded a drop out fee from me. It completely took me by surprise. It was incredibly frustrating because I had already brought up the extended version of my comic beforehand, and I had asked for the organizer's help in amending the ending to fit her 'completion' criteria. Therefore I told her she would have to wait for the 'drop out' fee that she demanded from me. During this period, she had proceeded to erase all of my involvement with SBCF'25. I believe this is a silencing PR tactic she had used with other artists with smaller voices.
As for my comic currently, it's already done and has been translated into Vietnamese, since I'm planning to sell the comic at my local conventions (Ho Chi Minh City). Please note that I had to abide by the Shortbox rule to not sell my comic anywhere during the fair period - October 2025 - that I had to skip participating in the only one artist alley we had in this city. Conventions in Vietnam are... very sparse especially for art and comics people, to put it lightly.
If this happened in gamedev, it'd be much easier for me to find outlets to speak out. I feel especially annoyed... and angry that artists have been such pushovers that we allow any level of micro or macro-aggressions to happen to us and just let the abuse continue. When I saw how much support and also pushback from the public when Horses gamedevs spoke out when their game wasn't allowed to be on Steam, I felt the same way: people turning a blind eye to Steam’s monopoly and their first reaction being "You'd have other vendors/publishers to pitch to". While SBCF is not a monopoly, it’s still the most well-known online comic fair in English. Artists have very little to no avenue to pitch their comic, especially if they’re new to the scene, and are not well informed on the existence of other online events if not already blacklisted by Zainab. She insists on claiming that others are plagiarizing her format, that they’re trying to become SBCF’s ‘competitors’ while in reality, these are organized by people who simply liked the idea of having more comics.
While sympathizing with Zainab's background - a brown Muslim woman who has experienced the worst parts of the comics industry - her mistreatment towards the artists who have built the reputation of the fair is, to me, intolerable.
I’m making this statement to warn other artists who are new to the comic scene, that if you also happen to get starstruck when being accepted to this seemingly prestigious fair. This is to remind long-time supporters of Shortbox the kind of behavior Zainab has behind the scenes, how she treats your favorite artists, and why so many of us decided to keep silent: it is to protect ourselves but also to protect the reputation of Shortbox.
Don’t apply to Shortbox. Support the artists directly when they post on their storefront. A lot of us posted our SBCF comics for free or at a discount after the fair was over. At the end of the day, we just want people to read our comics, and that’s good enough.
Did you hear the scientists have found a way to grow colored cotton? Thoughts?
It’s not a ‘scientists have found’ and much more ‘people have been already doing that for thousands of years and it’s just gaining more attention recently’
There’s actually been a load of vitriol leveled against folks who try to raise traditional colored cottons, because a lot of cotton growers don’t want the colored cottons cross-pollinating with their standard white cotton.
But anyway cotton can be grown in lovely natural shades of greens, reddish-brown ochres, and browns, all of which deepen with a good boil in water with a bit of washing soda thrown in.
The color obviously doesn’t fade or run, because it’s not dye. It’s the intrinsic color of the fiber itself.
I’m biased because I love the natural earth tones of many fibers, of course...browns, blacks, creams, copper-reds, ect...but I think they’re just gorgeous.
Scientists did not "just find out", and this is more of the same anti intellectual bs as the post that goes around claiming archaeologists were too stupid to know that hair could be sewn for elaborate styles.
Anyway, scientists DID figure out how to grow colored cotton. They genetically engineered it to be bright fuckin pink, and they didn't "just find out" about it, they already knew which is literally what inspired them to attempt the thing they just accomplished. Begging y'all to stop pretending that scientists don't know things, don't have interests, don't grow up in farming communities or have family who taught them this. Scientists are people. Do you seriously think people who use genetic engineering to make eco-friendly pink cotton don't know anything about textiles?
Anyway. Bright pink cotton without dyes, because science is awesome
Yes. CSIRO scientist Doctor Colleen MacMillan led the team that figured this out. They used tobacco plants for testing because of the genetic similarity. Basically if the tobacco leaves produced colors when injected with a bit of the experimental genetic material, the scientists on the team already understood that the color change would affect cotton bolls as well.
They grew bright red and bright yellow in a petri dish.
And yes, Doctor MacMillan knows lots of things. Here's a list of some of her publications.
@csirogram on Instagram
Additionally folks are researching how to create flame resistant cotton and black cotton. If a variety of black cotton becomes viable, it can stop a LOT of environmental damage caused by chemical manufacturing of black dye.
THIS. Every damn botanist I know, including myself, is at the least tangentially interested in fiber arts and indigenous methods behind things like that. Scientists have hobbies and we're all goddamn nerds so a lot of those hobbies are more niche. The anti-intellectualism is insane. I swear half of y'all think scientists are all evil cackling old men devoted to holding up colonial power systems. The work done by Dr. MacMillan is crazy cool and should be celebrated
I don’t even want to write a review of “Top Gun: Maverick” honestly, I’ve seen 1,500+ movies and I’ve never finished one that left me more depressed about the future of this country. I just want to stop thinking about it
The White House just shared a video using footage from “Top Gun: Maverick” right before real footage of US bombs hitting Iran. I am sorry to announce that I have been 100% vindicated in my views on this movie.
This is literally what people are talking about when they say AI will be used to mainstream widely held bigotry. LLMs are trained on frequency and probability -> straight relationships are more well represented in the dataset -> straight pronouns and terms become the "correct" normal.
This is a form of backdoor bigotry from both normative facts (there are more straight than gay relationships) and well represented bigoted beliefs (men are superior to women).
Combine this with the mass of people inclined to believe (and being encouraged to believe) that if AI says and does something it must be correct
the misogyny baked into LLMs is really quite horrific, we ran headfirst into this when trying to test out AI features at my job in early 2024. I work for a dating service and long story short there was a brief window when we were testing the capacity of an LLM to use the information on people's profiles to find matches for them and generate a brief bio that could be sent to the other person.
the problem that cropped up over and over and over again was that essentially all of these bios presented the women as looking for a heterosexual marriage with children regardless of any information on the profile that actively contradicted any of those things. even as the model got better and more streamlined at everything else, this persisted.
the features/pipeline we were testing out there i believe have been scrapped and the original idea is operating under a new name, but my team handles different stuff now. so i have no idea how the backend is being handled and if there's any AI in the pipeline, but i wouldn't be surprised if it was and this was still an issue.
wtf are you talking about, they didn't "deplete" the nature of their country, they cultivated their wilderness over centuries into some of the most idyllic pastoral landscapes in the entire world. And they did such a good job of it that the phrase "English countryside" is now synonymous with beauty and serenity and peacefulness. They didn't destroy their country's nature, they became its caretaker, they're right to be proud of it. All you're doing is pretending that the only kind of nature that should count is whatever is completely untouched by human hands.
The UK is literally ranked in the bottom 10% of most nature-depleted countries in the world. The 2023 State of Nature report shows just how dire the situation is. A third of UK bird species have declined since the 1990s, 97% of the UK's wildflower meadows have been lost since the 1930s, raw sewage is constantly being pumped into our rivers and seas with agricultural slurry also causing massive damage to rivers, three quarters of Britain's hedgehogs have been lost in the past 20 years and UK butterfly numbers are at their lowest ever, a sign of impending ecosystem collapse. Plus people in Britain are the most disconnected from nature in Europe.
Not to mention over populated deer destroying what little is left due to a lack of predators, 60 million non-native birds released for sport shooting every year, plus huge amounts of wildlife crime, including large numbers of birds of prey being shot/poisoned.
There is nothing beautiful about a sterile, ecologically damaged landscape that contains nothing but sheep and deer. Don't comment on something you clearly know nothing about. I live in England. I can see first hand just how dire the situation is.
Here is an article from NPR about it (May 22, 2026):
Carolina Milanesi, an independent technology analyst, said Google is trying to make its cash cow business — search — richer and more personalized, and it will make shopping easier. But there is a risk that users may have fewer choices about what to click.
"Right now it's: I ask a question, I get a bunch of answers and I feel that I'm in control as to which answer I take, or if I'm looking for something, which product I'm going to end up buying. That is going to be less so going forward," she said.
Milanesi envisions AI-enabled search and agents proposing products to consumers — perhaps even those they have requested — but with less clarity or choice around where it's coming from.
"If you're going to say: 'I want a pair of Jordans, go find them,' you're not necessarily sure what steps have been taken and whether the AI has used a source or a store that was paid for and therefore came up in the search results," she said, "or if AI actually went and did their due diligence and picked the best for me as a customer."
And here's one from Time magazine (May 20, 2026):
While Google already has “AI Mode,” the company will now power the whole search bar through its new Gemini 3.5 Flash model.
Instead of the classic list of blue links, Google Search will now also generate a custom page with an AI-generated summary of what you’re searching about, which will then trigger a conversation with AI Mode on the main page, allowing users to ask follow-up questions—similar to the kind of layout you would see when opening ChatGPT.
And a little more from Time's article on how this may affect the websites that we are trying to search for:
When Google first started implementing AI-assisted results, news publishers warned of “catastrophic” impacts on the industry, much of which relies on Google search to drive users to their websites.
Last year, news websites saw significant traffic declines as chatbots increasingly replaced Google search as the primary way to find sites and ask questions.
Small businesses also noted drops in traffic to their sites from Google, which has traditionally delivered customers.
Lily Ray, vice president of SEO strategy & research at Amsive, a digital marketing agency, warned as early as last year that Google’s planned changes to search are “going to have a devastating impact on the Internet.”
“It will severely cut into the main source of revenue for most publishers and it will disincentivize content creators who rely on organic search traffic, which is millions of websites, maybe more,” she told Technology Magazine.
random PSA, I know a lot of people use duckduckgo as a Google alternative search engine, but it always kind of annoyed me when I was using it because it felt like No Name Brand Google
I have switched to using Startpage.com and vastly prefer it. for one thing, instead of displaying an "AI summary" at the top of the search results (unless you turn it off, yes I know), it displays the first paragraph of the Wikipedia article, with link, whenever it finds one that's relevant.
also a waaayyyyy better sense of design than duckduckgo
also private, European based, least annoying search I've used lately (RIP old "don't be evil" Google)
i have one of those, scraped from multiple different rec posts:
Search Engines
Infinity Search is an alternative search engine with a special focus on privacy
DuckDuckGo is a popular search engine for those who value their privacy and are put off by the thought of their every query being tracked and logged. Uses bangs, ![site] for in-page search (sells your data to microsoft and draws from fucking bing)
WolframAlpha is a privately owned search engine that allows you to “compute expert-level answers using Wolfram’s breakthrough algorithms, knowledgebase, and AI technology.” A data search engine.
Boardreader is a search engine for forums and message boards. It allows you to search forums and then filter down results by date and language.
Based in France, Qwant is a privacy-based search engine that won’t record your searches or use your personal details for advertising. Uses “&” as a bang search.
Another privacy-based search engine is Search Encrypt, which uses local encryption to ensure that users’ identifiable information cannot be tracked. Metasearch across multiple engines.
Offering unbiased results from several sources, SearX is a metasearch engine that aims to present a free, decentralized view of the internet. Can be self-hosted.
Gibiru’s tagline is “Unfiltered private search” and that’s exactly what it offers. Requires AnonymoX Firefox add-on for privacy.
Disconnect allows you to conduct anonymous searches through a search engine of your choice.
Swisscows provides fully encrypted searches to protect your privacy and security. Built-in violence/porn filter cannot be overridden.
MetaGer offers “Privacy Protected Search & Find” through its anonymised search. A plugin will allow it to be made a default.
Gigablast is a private search engine that indexes millions of websites and servers real-time information without tracking your data, keeping you hidden from marketers and spammers. Variety of filtration and refinement options for searching.
Oscobo is a search engine that protects your privacy while you search the web. By not using any third-party tools or scripts, your data is protected from hacking and misuse. Has a Chrome extension to allow use in toolbar.
https://search.marginalia.nu/ an independent DIY search engine that focuses on non-commercial content, and attempts to show you sites you perhaps weren't aware of in favor of the sort of sites you probably already knew existed. Use old-school searching rather than query-based for the best results.
https://www.mojeek.com/
https://wiby.me/ - It’s goal is to index as many personalized websites as possible, and NOT commercial sites.
https://4get.ca/ it works a lot like SearX, but honestly better. It doesn’t have its own index, but pulls from many others. I think it’s the best for research, since it allows you to search for answers from different indexes, is easy to configure, add free, and avoids censorship as much as it can.
https://www.searchenginemap.com/ for more on how search engines relate to each other.
https://yep.com/ is a crawler
https://www.etools.ch/ retrieves from Google, Mojeek, Bing, and Yandex, like Searx
https://www.dogpile.com/
https://searxng.org/ (next gen Searx)
https://luxxle.com/ - possibly conservative?
https://presearch.com/ - good for academic?
https://kagi.com/smallweb - free/randomised Kagi.
Other Searchers
www.refseek.com - Academic Resource Search. More than a billion sources: encyclopedia, monographies, magazines.
www.worldcat.org - a search for the contents of 20 thousand worldwide libraries. Find out where lies the nearest rare book you need.
https://link.springer.com - access to more than 10 million scientific documents: books, articles, research protocols.
www.bioline.org.br is a library of scientific bioscience journals published in developing countries.
http://repec.org - volunteers from 102 countries have collected almost 4 million publications on economics and related science.
www.science.gov is an American state search engine on 2200+ scientific sites. More than 200 million articles are indexed.
www.base-search.net is one of the most powerful researches on academic studies texts. More than 100 million scientific documents, 70% of them are free.https://cosine.club/ is an electronic music similarity search engine
404 has been knocking it out of the park since they started. Please support their original reporting on this! If you subscribe to nothing else I highly recommend them. Their podcast is great too.
Planning documents for "Scout" say the plan is to "make people addicted" to the tool before adding new features.
"While those working at private companies can at least earn a little money, they face possible punishment if they refuse, from being denied family visits to being sent to higher-security prisons, which are so dangerous that the federal government filed a lawsuit four years ago that remains pending [note: article is from 2024], calling the treatment of prisoners unconstitutional.
Though they make at least $7.25 an hour, the state siphons 40% off the top of all wages and also levies fees, including $5 a day for rides to their jobs and $15 a month for laundry.
Turning down work can jeopardize chances of early release in a state that last year granted parole to only 8% of eligible prisoners — an all-time low, and among the worst rates nationwide — though that number more than doubled this year after public outcry."
No state has a longer, more profit-driven history of contracting prisoners out to private companies than Alabama.
i know things are hella grim in the nsfw/kink art circles especially in the last year --
but I'm hearing there's a NSFW-friendly ko-fi alternative built on atproto that's actively in the works, and being vetted by lawyers right now. as torrent-princess (OP) says, you should be able to swap out payment processors while keeping your account intact. this matters since even if stripe removes support, you'll still have a shop and all of your links intact. (ATproto is an infrastructure that bsky is built on, but is far bigger than bsky with far more opportunities.)
additionally, the Free Speech Coalition is working on a credit union specifically for adult work (including kink art) - here's the link so you can add your interest & support. Since this will be built by sex workers, there'll be far less risk of being debanked for spurious and puritanical reasons.
on a domain TLD level, there's an initiative here for a .furry domain built from the ground up by seasoned furries; it's unclear whether they'll support NSFW, but it's yet another promising turn of events for a group that's been similarly affected by censorship.
there are friends and allies out there helping to build a working parallel infrastructure. keep being vocal, keep supporting these initiatives when it's possible, and keep supporting your nsfw/kink artists. ♥
maybe y'all didn't notice but fat people who don't hate ourselves sure did notice that people were obsessed with shitting on fat people in the late 90s and early 2000s (conservative political time) and now are again (fascist political time), coincidentally while the market for weight loss has become a 90 billion dollar industry due to glp1s.
you are not immune to propaganda. it makes some people a whole hell of a lot of money for you to hate fat people and fear becoming (or staying, I think like 70% or something of the US is fat) one of us.
a lot of the fearmongering over fatness comes from studies directly funded by the weight loss industry...i think people don't really realize or think about the fact that research can absolutely be influenced and skewed by its funding. there is also research that shows that an amount of the negative health outcomes for fat people come from anti-fat bias. if you go to the doctor with concerns and the doctor simply tells you to lose weight, your problem is neglected and you may not even bother going to the doctor with the next problem.
every fat person you know for the most part probably has a story like this, of medical neglect. many of the stories i've heard personally are when the complaint or the doctor wasn't related at all, like being told to lose weight at the ear nose and throat doctor or at the dentist. it's straight up just bias. it's such a thing that in the show Shrill it's portrayed, when Aidy Bryant goes to the gynecologist and her doctor suggests she get gastric bypass.
the studies on health and fatness are simply not that black and white and there is basically no research that shows that more than an incredibly tiny minority of people can lose weight and keep it off for more than like 2 years. bodies have set points that they gravitate towards, it's not a personal failure. this also is how the weight loss industry succeeds so well - repeat customers.
some of the harm associated with fatness is also due to weight cycling, which is very hard on your body and is even worse if you get off a GLP1, which according to a recent study causes weight to be regained at a rate that is 4x faster than without taking a GLP1.
you don't have to hate yourself. you don't have to hate other people for their body type either. it makes me so sad to see the thinspo tag going around again in 2026 a lot like it was back in the day.
A study spanning almost four decades and involving more than 100,000 adults in Denmark found that those with an 'overweight' body mass index
there's so much crazy shit once you go down the rabbit hole. for example, BMI was not invented by anyone with a medical background. it was never meant to measure individual health.
I think the piece of information that most firmly changed my mindset is understanding that sustained weight loss is essentially impossible. It might be true (and likely is true in at least some cases) that weight is in some ways a protective health factor; health issues cause weight gain and not vice-versa; trying to lose weight and fat stigma both cause health issues rather than fatness itself. But none of those arguments even matter at all if there is no intervention that has ever been found to substantially reduce weight for more than like, two years, after which everyone gains it all back and suffers new health problems for having starved themselves. Even if it's true that weight itself causes health problems, every doctor telling you to lose weight might as well be telling you "have you tried getting younger? Aging is really bad for your health."
Text of tweet under the cut because it is loooong.
But... Stochastic Parrots.
Timnit Gebru was fired from Google in December 2020 for refusing to retract a research paper, and every single warning that paper made about large language models has now happened at a scale the industry spent 4 years trying to make people forget about.
Her name is Timnit Gebru.
She co-led the Ethical AI team at Google. She co-wrote a paper called "On the Dangers of Stochastic Parrots" with Emily Bender at the University of Washington and two other researchers. The paper was 14 pages long. It was submitted to a top AI ethics conference. And it was the reason Google decided that one of the most senior Black women in AI research could no longer work there.
The story Google told publicly was that she resigned. The story she told, confirmed by 2,695 of her colleagues in an open letter, was that she was fired by email while on vacation because she refused to either retract the paper or remove her name from it.
The paper had not even been published yet.
Here is what she actually wrote, and why every prediction inside it has now come true.
The first warning was about scale itself. Bender and Gebru argued that training ever-larger models on ever-larger scrapes of the internet would produce systems that appeared fluent but had no actual understanding of language. They called these systems stochastic parrots because they would repeat patterns from training data with statistical confidence and zero comprehension. The paper predicted that this apparent intelligence would fool both users and developers into trusting outputs that were structurally incapable of being reliable.
This was 2020. GPT-3 had just come out. The paper predicted the hallucination problem before anyone had a word for it.
The second warning was about bias amplification. The paper documented in detail that internet-scale training data contains systematic overrepresentation of dominant viewpoints and underrepresentation of marginalized ones. The models would not just absorb this bias. They would amplify it, because the optimization process rewards confident outputs, and confidence in language patterns tracks frequency in the training set.
The prediction was that hiring tools built on these models would discriminate against women. That healthcare triage tools would underperform on Black patients. That loan approval systems would entrench inequality while presenting their decisions as neutral algorithmic judgment.
Every one of those things has now been documented in deployment.
Amazon's hiring algorithm penalized resumes that contained the word "women" in any context. Healthcare risk scoring algorithms used by major US hospitals were found to systematically underestimate the medical needs of Black patients. Apple Card's credit algorithm gave wives credit lines 10x lower than their husbands for the same financial profile.
The third warning was about environmental cost. The paper calculated that training a single large language model produced emissions equivalent to the lifetime output of 5 cars. The prediction was that the race to scale would create an environmental footprint that would eventually rival entire industries.
In 2024, Google's emissions were up 48% from 2019, and the company explicitly blamed AI infrastructure. Microsoft's were up 29%, same reason. Both companies have now quietly abandoned the climate commitments they were publicly celebrating the year Gebru was fired.
The fourth warning was about documentation. The paper argued that the training datasets being assembled were too large for anyone to actually audit. Nobody at Google, OpenAI, Meta, or any other lab could tell you with confidence what was in the data their models were trained on. This was not a temporary problem to be solved later. It was a permanent feature of the approach.
In 2023, researchers discovered that the LAION-5B dataset, used to train Stable Diffusion and other major image models, contained thousands of images of child sexual abuse material. The companies that had trained on the dataset had no way of knowing. The paper predicted that category of failure 3 years before it was found.
The fifth warning was the one Google cared about most.
Bender and Gebru argued that the deployment of these systems would centralize linguistic and cultural power in the hands of the small number of companies that could afford to train them. The internet would become a place where the dominant voice was a statistical average of dominant voices, presented as a neutral assistant. Languages underrepresented in the training data would degrade over time as more web content was generated by these systems and fed back into the next training run.
This is now happening in real time. A 2024 study found that 57% of new web content in English is AI-generated or AI-assisted. Researchers studying low-resource languages have documented active degradation in translation quality, because the synthetic content fed back into training is itself worse in those languages.
The paper Google fired her for predicted the model collapse problem before model collapse had a name.
The mechanism behind why this all happened is the part of her work that nobody quotes.
Gebru's argument was not that AI is dangerous in some abstract sci-fi sense. Her argument was that AI is dangerous in a very specific structural sense. The technology was being built by a small group of researchers who shared similar backgrounds, worked at similar companies, and were rewarded for shipping products faster than competitors. The incentive structure made it impossible for safety, ethics, and bias concerns to slow anything down. Anyone inside the system who raised those concerns was either ignored, sidelined, or removed.
She was making that argument from inside Google.
Then Google proved her right by removing her.
The team Google had built to make sure their AI was safe was dismantled in 90 days because they did the job they had been hired to do. Margaret Mitchell, the other co-lead of the Ethical AI team, was fired two months after Gebru for searching through her own emails for evidence of how Gebru had been treated.
Gebru did not stop. She founded DAIR, the Distributed AI Research Institute, in 2021. The mission is to do AI research outside the control of the companies that have a financial interest in not hearing the answers.
Every prediction in the Stochastic Parrots paper has now been validated by deployment. Hallucinations are an industry-wide problem the largest labs cannot solve. Bias amplification has been documented in hiring, healthcare, lending, and criminal justice. Environmental costs are larger than entire small countries. Training data audits remain impossible. Model collapse is an active research crisis at every major lab.
The question worth sitting with is the one almost no one in the industry will say out loud.
Every researcher with the technical credibility to call out these problems watched what happened to her in December 2020 and made a calculation about their own career. The number of people willing to speak publicly about safety and ethics issues inside the major AI labs collapsed after that firing and has not recovered.
The researcher Google fired for warning about exactly what is now happening was right.
The company that fired her is now the second-largest deployer of the technology she warned about.
And the people inside that company who agree with her are not allowed to say so.
pared down to the bone @chaoticgoodrandom - Tumblr Blog | Tumgag
