Nov 15, 2026, 12:30 PMa local queer org for young girls (13-28) is hosting a pro "AI" 'informational' afternoon and besides all the general reasons against LLMs i need ressources on individual LLM providers (Grok, ChatGPT, Gemini, ...) and why *especially* young girls shouldn't use them.
I will do everything in my power to shift the discourse away from "let's write fun stories with "chatty", look how totally not queerphobic the answers are!" to real world implications of fash-tech.
Nov 15, 2026, 12:44 PM@CIMB4 for grok, i'd recommend bringing up how bad actors used it for CSAM and deepfake porn, etc.
for chatgpt, i'd talk about cases of suicide in teens caused by worsening mental health and encouragement along with medical "advice" that has killed people from chatgpt.
you could also use the same for gemini, but i'm not sure if i'm misremembering, etc.
while the effects on developing brains of LLM use aren’t really that studied that I know of, given there’s evidence (besides how many LLM bosters are total dipshits) of cognitive decline in adults from LLM use (https://www.media.mit.edu/publications/your-brain-on-chatgpt/), I imagine the effects on young minds are pretty fucking bad
also the tie between LLMs and youth suicides is non-negligible
for specific models uhhhhh
facebook’s models specifically have a tendency to engage minors in lewd conversations
grok has a tendency to generate CSAM when asked, and push right wing politics sometimes
Nov 15, 2026, 1:34 PM
Nov 15, 2026, 12:56 PMhttps://codeberg.org/ethical-foss/ai-free-foss-alternatives/src/branch/main/why_not_llms.md
The above has a collection of links and articles on "Why not LLMs"
It has a nice assortment of articles and domains including:
* Legal Ramifications
* Stolen Training Data
* Environmental Impact
* Labor Issues
* Code Quality Issues
* Health and Safety Issues
* Effect on Hardware Prices
@CIMB4 LLM use in general results in cognitive decline. I wouldn't let my teens anywhere near ANY of them.
@CIMB4 are the numerous suicides and mental health crises exacerbated by chatbot use counting
@CIMB4
chatgpt was specifically implicated in the tumbler ridge school shooting https://www.cbc.ca/news/canada/british-columbia/tumbler-ridge-shooting-open-ai-lawsuits-9.7328382
also grok has been involved in CSAM stuff but others seem to have mentioned that already
Nov 15, 2026, 2:27 PM@h33h33wuffi3@plush.city @CIMB4@norden.social a side note that as a victim of childhood sexual abuse i absolutely despise people using the term "CSAM" to describe ai generated stuff. https://moth.monster/blog/artificial-iconoclasm/ the distinction between real and fake matters more than ever these days.
AI being a force multiplier for sexual harassment is a real problem, one worth taking seriously. along with all sexual harassment. weird how everyone wants to blame little kids for not protecting themselves from sexual harassment enough, instead of the people actually doing it... that's out of scope though
Nov 15, 2026, 2:46 PM@h33h33wuffi3@plush.city @CIMB4@norden.social frankly i think we should stop focusing on the images, and start focusing on the actions. if one truly needs a term for it i think "deepfake" is a better name since it focuses on the part that the images are FAKE, which is important.
if someone's using AI generated images to sexually harass someone, that's just called sexual harassment. the methods they are using don't matter as much as the fact that they decided to sexually harass someone using whatever tools they had.
stop blaming the tools, and stop worshipping icons.
Nov 15, 2026, 2:57 PM@eri @h33h33wuffi3 if i may carefully interject: CSAM to me(!), means 'material' or media stemming from CSA. when there's a real photograph of a child in lingery, one can safely assume the prerequisite of that photograph's existance was CSA (forcing the child to pose and expose themselves etc.)
fake imagine generation doesn't remove the CSA from the CSAM because the model was trained on 'real' CSAM.
@CIMB4 I'm unsure of specific resources but I know with grok specifically, it's been used heavily to create csam revenge porn of girls using like, just selfies.
Nov 15, 2026, 2:22 PM@CIMB4@norden.social there's really no reason to specifically single out any providers as being better or worse because it's all the same underlying tech. i think any argument you're making should be very clear about this; it's a digital yes man who always agrees with you, ignoring obvious warning signs.
Nov 15, 2026, 2:45 PM@eri Yes agreed, however i specifically want examples for individual services so they can't argue "well but that's about service a and we're usinf service b!". i've already collected more general info and need help finding individual cases.
-like if i bring up grok's csam or google/gemini's finnish forest planing they'd say "but chatgpt didn't!"
Nov 15, 2026, 2:24 PM@CIMB4 I think a /very careful/ mention of the potential for AI Psychosis -- in the clinical/near-clinical "impact ability to engage with reality" sense, not just the "buys into the hype and uses/pushes it heavily" sense -- in how it, from the implementation choices made about its interface and trained into it slightly-more-deeply, has a pattern of,
for people with issues of psychosis or with some potential to develop psychosis,
exacerbate those and trigger/push towards breaks with reality in those cases. If you find psychs talking about this they'll probably have stats on the base rate of that in society (and so, base rate of people vulnerable in this way), which is higher than people think,
and you have a responsibility to do programming that is safe and not risky for all your students.
(And, as of yet, no meaningful changes have been impemented to make them safer in this way, or socially to proactively approach this harm.)
Nov 15, 2026, 2:36 PM@CIMB4 (The "carefully" I was thinking was -- well, (a) not stigmitizing psychosis/people who experience it, but also (b) making sure not to come off as claiming(/accusing) them of being harmed in this way, which I think wouldn't go well and would undermine the point.)
(The way I phrased this was intentionally more able-to-be-conciliatory towards the main LLM claims. I personally think that the "stochastic generation of maximally plausible language" model as a whole is structurally tilted towards this sort of risk as a whole (especially with the "we'll put ad-hoc shims for common interactions"), and isn't likely resolveable by local changes, and needs a paradigm shift. And that, in the "this is fash-tech broadly", sure rhymes HARD with the attack on the human ability to observe and validate consensus reality that is part of the fascist political project.
But I think "in addition, this angle ALSO" might reach /them/ more than "this is an extension of other issues".)
Nov 15, 2026, 2:49 PM@gaditb i won't lie, my ME/CFS made it difficult to parse your run-on sentences but i think i got ur memo and feel like i've heard this before. do you remember any source? i want to be ay credible as possible so the organizer can't stamp me as "overly concerned" or "fear mongerer"