in a way john watson is a fantasy (what if you had this brilliant enigmatic friend and what if he liked you in particular and what if he offered you the excitement of youth and adventures and a way out of boring society life and all without having to actually give up your status as a gentleman so you could have the best of both worlds) and in a way sherlock holmes is a fantasy (what if someone never got tired of you despite your various strange habits and mood swings and instead of simply tolerating you they genuinely liked you and what if you didn’t have to live alone forever and what if you never had to give up doing the things you love) and of course there’s the most fantastical part of it all (what if you could afford london housing prices)
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.
YOUNG BROWN AND BLACK GIRLS: IGNORE THE WHITE INFLUENCER TELLING YOU TO GET BOTOX OR A NOSE JOB!!!!! YOUR FACE IS BEAUTIFUL AND YOU ARE WORTHY OF LOVE JUST AS YOU ARE ‼️‼️‼️‼️‼️‼️
Me: "Damn people are REALLY BAD at knowing when to tag their eyestrain art/images...either that or they just don't care about photosenitive epileptic people like me. I feel really sad now."
Person: "But Allison, what if they just don't know or understand what qualifies as eyestrain and what doesn't?"
Me: "You know what? That could be a factor...While it is always better to be safe rather than sorry (so YES people should always tag eyestrain even if they're unsure if it "counts" or not) maybe you've got a point?"
Anyways! HERE'S YOUR HANDY GUIDE TO WHAT CAN COUNT AS EYESTRAIN! I'm pulling this straight from the Artfight rules page about what needs to be labeled and filtered as eyestrain because it's VERY helpful and VERY accurate! I also know not everybody has an AF account and might not always have access to this handy guide, and this is an important resource; That's why I'm sharing it here! (under the cut)
PLEASE TAKE THIS SERIOUSLY!!! THIS IS ABOUT THE HEALTH AND SAFETY OF OTHERS!!!
Full eyestrain AF page link
"But Allison! How were you able to screenshot that example if you're so sensitive to eyestrain?"
I dimmed the HELL out of my computer screen and looked away while taking the screenshot and did the same when putting it into this post, that's how lol. BUT YEAH ANYWAYS!!! Once again:
PLEASE TAKE THIS SERIOUSLY!!! THIS IS ABOUT THE HEALTH AND SAFETY OF OTHERS!!!
For a second I didn’t realize it meant “high” as in a stoner–I thought “High Geologist” was like a rank of geologist or something and he was insulted you would challenge him to naming stones
i don't remember them playing mahjong but they do other old man things like going to the wet market together and drinking soup and taking walks. anyway go watch suk suk / twilight's kiss
What do you mean “chat” is now referring to ChatGPT and not twitch chat? What? What? What the fuck? No?
When I address chat I am speaking to a presumed Greek chorus of real human people shitposting on their lunch break, not a machine that devours lakes to covert electricity into slop.
If you are unfamiliar with Tofu, please read this post or watch this playlist about the snake who hates Markiplier.
tl;dr- under the Markiplier gif.
From the videos you can see she tries to figure out what is upsetting Tofu, and found that it's not just streamers, or just his voice, or the stream. This specific snake
He does not like Markiplier specifically which is demonstrated by him squaring up with Mark, striking, and tail buzzing at him. He recognizes Mark specifically as a scary threat.
Animals have been shown to recognize human faces, and Tofu's reaction to Markiplier's face (and only his face as shown by the printout) is strong evidence that he recognizes Mark's face and differentiates him from other faces. With crows, they've been shown to hold grudges and have that grudge extend to other crows.
So Tofu has a negative association with Mark, and specifically Mark. The question is why? My theory is: perceived threat after a perceived dangerous situation. Some have theorized Tofu was mistreated by someone who looked like Mark- if they were while in his keeper's care she'd know. If it was his breeder, he'd more likely be afraid of all humans and need to first get over this negative perception (basing this on how Sakura reacted with massive trauma that took years for her to overcome, and to this day she's still jumpy. A breeder doesn't socialize their snakes so they'd only know humans to be abusive before being adopted, and his current keeper would very much notice that terrified trauma behavior to herself.)
One thing important to know about snakes is they are not only capable of learning from each other, but are protective of each other. I've seen this with my girls interacting with each other (yes snakes have friends and remember their family), learning from watching the other and copying her actions, as well as behavior I've observed on the Rattlecam with wild snakes. Adult snakes teach younger snakes, as well as are protective of them- even if they aren't their own babies. A female was shown on the rattlecam watching over a group of slitherlings that were obviously not her offspring as she was heavily pregnant. IIRC that same female ushered babies to safety after a hawk attack before worrying about her own safety. I've seen it in person as well with my girls, who have shielded each other from perceived danger, and scoria also squared up to something scary in the hall as though she were protecting me. And when it left she immediately went back to being relaxed and cuddly and happy. Snakes have friends that are snakes. It's an outdated belief snakes cannot bond with their keepers, as many do and seek them out for attention. These observations of snake behavior are part of what play into my theory basis.
So what could Mark himself have done? If you watch streamers you'll know Markiplier is known for playing jumpscare games with his FNAF let's plays being some of his most well known. Tofu's owner knows Markiplier, and is actively watching Markiplier with Tofu. More than likely she regularly watches Mark, has watched jumpscare videos, and had Tofu around while doing this.
It's a reasonable assumption. And if Tofu's keeper reacts to Mark's jump scare videos by flinching, looking frightened, or making frightened sounds, it's quite likely Tofu's sees Markiplier is involved every time... and blames Mark. Seeing his keeper get startled, Tofu doesn't understand it's a fun video or that humans enjoy being jumpscared (honestly I don't think I could explain it to him either) so he's learning from his human that Markiplier is a scary threat.
And from Tofu's point of view? He's right! Markiplier is a regular threat that not only (probably) scared his human, but regularly appears to stare him down and even went after him in his own home! And when his keeper scolds him? It's likely the same misunderstanding that dogs have when their owners yell at them to be quiet- they think the human has joined in! Clearly his keeper is joining him in being upset at Mark!
So in Tofu's mind? Mark is a diabolical threat to his family's safety that he must defend them against, and they are united with their keeper in their efforts to drive away that scary scary guy. You are very brave Tofu, and we are all very proud of you.
The way that most of Conan Doyle’s Sherlock Holmes stories’ most horrible villains are rich dudes that are abusive to women, in a time such as the 1880’s, compels me.
Yup, there’s a huge number of times where Sherlock Holmes is the ONLY person to take a young woman’s complaint or worry seriously and finds out someone is up to some serious evil. Holmes also shows a lot of compassion and empathy with the victims over and over again. (This is why I find “Secretly a woman” or “Trans” Holmes headcanons much more convincing than “sociopath” Holmes.)
I am never going to shut up about how much I specifically love The Adventure of The Copper Beeches because it is literally Sherlock Holmes listening to a young lady he does not know except as a potential client, agreeing with her that a potential job she has interviewed for that she thinks is SUPER SKETCHY is, indeed, sketchy as fuck and when she says she’s probably gonna take the job anyways because the money is good and she needs it going “OKAY I GUESS but for the love of god please write to us so we know you’re okay we will literally drop everything and jump on a train if you want us to”.
The job turns out to indeed be sketchy as fuck, she writes to them, Holmes and Watson drop everything and jump on a train when she asks them to. I read this story for the first time when I was twelve and it made a HUGE impression.
This is also the basis for a lot of speculation about Holmes’ family life. The idea that he has been a victim of abuse, or his mother was abused (or even murdered by his father.) There’s definitely SOMETHING that makes him very aware of how dangerous isolated families can be, and the dark things that can happen behind closed doors. Plus, of course, the motivation to devote himself to stopping crime. And yes, so much of it is of the personal type.
dude see this is one aspect of the original books i NEVER understand why modern remakes (cough cough) don’t go all in on. Like, in the 21th c we HAVE all the dumb forensic shit that made Victorian Holmes stand out, but we STILL DON’T HAVE uh….you know, compassion for women and minorities, or the willingness to believe them, adequate community support for domestic violence or hate crimes, etc. etc. which you’d think is exactly where a renegade consulting detective would come in handy. A good modern day Sherlock Holmes remake, instead of trying to convince us that Holmes is some super genius for being better than fingerprint analysis or whatever, could have him just be…a good person who helps out people the police can’t and won’t help. There you go. That’s how to write a relevant modern Holmes.
One thing that annoys me is how much the BBC version of Sherlock (and the fandom around it) focus on police cases or cold cases. In the stories, Holmes’ bread and butter cases had fuck-all to do with the police and in a few stories, he actively works around/against them, or outright lies to them. Of the many, many things I wish that show had done differently, this is one is particularly obnoxious since it’s such a gimme.
There were very few actual murder cases in the Canon, and Holmes handled them either one of two ways:
Option one: The murder victim was innocent while the killer was an abusive bastard, see Speckled Band. Conclusion, arrest and have the killer charged (Or in the case of Speckled Band, indirectly murder him yourself then shrug and go home)
Option two: The victim was murdered to protect someone that the victim was abusing, or for vengeance, see Boscombe Valley, Devil’s Foot, Abbey Grange. Conclusion, Oops, I don’t know who the killer is, I am suddenly incompetent, oh look a pheasant.
#my favorite murder in holmes canon#is when they straight up witness a lady murder her blackmailer#do nothing except destroy his other blackmail material#and then straight up lie to lestrade about it#sherlock holmes#more of this in modern adaptations pls (via @cactusspatz )
Let’s not forget the time Holmes helps a young woman who’s being catfished by her own stepfather to steal her inheritance, and when the villain sneers that the law can’t touch him, Holmes grabs a horsewhip out of sheerest chivalry.
