iiinova — where your digital self keeps evolving

Reposted by @amatheac.bsky.social

Ruby, pretty ape machine @anontranshoe.bsky.social · 03/10/2026

just having a mild freakout cuz this is due in 10 days and I'm at $27/$1234; please help a household of 4 trans ppl retain their electric 🙏💚

11626

amatheac.bsky.social @amatheac.bsky.social · 03/10/2026

Can I offer you an excavator? 'cause you seem to be committed to digging yourself into a deeper hole with this tantrum ... just stop flailing and sit with the fact you're out of your depth here. Sheesh.

020

amatheac.bsky.social @amatheac.bsky.social · 03/10/2026

IMO it makes a difference I am having this conversation in public. i) onlookers benefit from understanding there are flaws + what those are (stuff to look up) and ii) I can simultaneously turn it into a draft for a more permanent rebuttal if need be, but yeah.

130

amatheac.bsky.social @amatheac.bsky.social · 02/10/2026

More - actually in an FB group having a back and forth with Artemis re: stats analyses being egregiously flawed. Reposting with alt text.

The thresholds you need to be using for p values in the analysis of function words is p < 0.00625 (Bonferroni correction for the Type I FWER , formula is alpha / number of tests). You can also generate false discovery rate corrected p values using Benjamini-Hochberg correction , but if you specify a threshold of 0.05 5% of your significant hits will be false positives.
I'm also not confident at all about your machine learning analysis re: telltale signs of Claude Opus outputs because of how similar the distributions are between pasted text and typed text. Either pasted vs typed text is not a germane distinction, or trying to quantify how "Claudelike" the writing is isn't capturing an underlying phenomenon. It can't be both.
With reference to the modelling strategy here "We then trained a classical topic model (an old-school machine learning model) on each of the groups in the known-human and known-machine texts, and compared the results of both sets to Reed’s actual writing" you need to be using a nested validation strategy - split the data into a series of training and test sets ; fit a classifier to set apart 100% machine written and 100% human text within the training set specifically , and then calculate precision-recall stats + Cohen's Kappa in the held out test set. If you are training on the whole set any performance measures, even if you calculate them using cross-validation, are not appropriate because I guarantee they represent model overfit. You can then weight the class probabilities output by each of the models fit by their Cohen's Kappa in held-out-test sets to give you an ensemble score you can use to classify additional unlabeled data.

171

amatheac.bsky.social @amatheac.bsky.social · 02/10/2026

Sexual abuse in frat settings may be so much more common than otherwise believed. pubmed.ncbi.nlm.nih.gov/31984858/

010

amatheac.bsky.social @amatheac.bsky.social · 02/10/2026

Re: that clownish hit-piece on Erin Reed from the Needle. There are some egregious holes in their statistical "analysis". It took me less than 30 minutes skimming it to pick these ones out. If I have the time or the inclination I might do a deep dive.

$Amathea Chakravarthy · The "analysis" in the paper is full of holes. You do not have any reliable estimates for the area under the PR curve and Cohen's Kappa for accurately distinguishing AI-generated text from completely human generated text using your approach. Your examples are cherry picked. You need to have controls where you deliberately create spliced AI-human text at various mixing fractions and show you can reliably distinguish these synthetic controls from the fully human equivalents. You also need to show any patterns you observe for Erin Reed are unique to her when compared to a broad set of journalistic writing on similar events. Just looking at S.Baum is a piss-poor control. You refer to a paper that talks about Stylometric identification of AI-generated text with "98% accuracy". Accuracy is not a rigorous metric at all because it is remarkably heavily biased by class imbalance. It should be trivial to observe that if I have 5% of one group , 95% of the other, I can build a classifier that calls everything the majority group and it will still be 95% accurate despite not being capable of any classification at all. It is not reassuring that you give this metric any credence. For the analysis looking at rates / 1000 words the normalisation is iffy because a shift in the overall occurrence of these categories of words can also produce differences - it is a confounded variable, you need to be normalising to rate per 1000 demonstratives , rate per 1000 modals et cetera. The data exclusion in the intervening period is iffy given that your claim involves a continuous temporal trend (read - https://pubmed.ncbi.nlm.nih.gov/29933352/ ). You perform multiple tests and don't correct for false discovery rate or the Type I family-wise error rate here. All of this is basic statistics.$

2285

Reposted by @amatheac.bsky.social

Doc Vivi Leandra【 VTC 】- Real-Life Scientist VTuber 🧬🖥️ @docvivileandra.bsky.social · 29/09/2026

Katra's a good friend, and I'd appreciate if any of the lab rats could help her out!

02517

amatheac.bsky.social @amatheac.bsky.social · 28/09/2026

Luetke et al documented extremely high rates of male-on-male sexual violence victimisation within fraternities. pubmed.ncbi.nlm.nih.gov/31984858/

000

amatheac.bsky.social @amatheac.bsky.social · 28/09/2026

Re: Cornell and the incredulity around why nobody else in the fraternity did anything to stop the rapes ... fraternities are hotbeds of sexual violence ; men in them are 3x more likely to perpetrate rape than non-frat men. Even the rate of male-on-male SA is much higher within frats. Links follow...

100

amatheac.bsky.social @amatheac.bsky.social · 19/09/2026

If AI kills a lot of people it will be because of idiots like these military leaders. FFS.

000

amatheac.bsky.social @amatheac.bsky.social · 18/09/2026

What I want to know is if this species also purrs/makes biscuits.

010

amatheac.bsky.social @amatheac.bsky.social · 17/09/2026

Even with defined neurological systems like grid cells, resemblance to actual biology requires ad hoc model retuning until you get output patterns that resemble biological patterns. Obviously this completely skewers ideas of intrinsic mappability between artificial neural nets and biological ones.

010

amatheac.bsky.social @amatheac.bsky.social · 17/09/2026

She's right, you utter dolt. There is a reason that language-focused neural networks that embed inputs as tokens that can be passed through multilayered neural nets have to be reoptimised ("finetuned") for specific applications, often using human feedback to reinforce specific stochastic outputs.

020

amatheac.bsky.social @amatheac.bsky.social · 17/09/2026

Calling this class of models neural networks was a huge mistake, good lordy.

030

amatheac.bsky.social @amatheac.bsky.social · 17/09/2026

TIL matrices are "thoughts". So I was torturing sentient matrices whenever I got a problem wrong in my middle school matrix multiplication airthmetic, got it. Fucking hell lmfao.

050

amatheac.bsky.social @amatheac.bsky.social · 17/09/2026

re: frontier LLM debate, IMO many are missing that the potential for harmful,unpredictable behaviour/model use doesn't hinge on symmetry with usefulness. Dangerous =/= smart, inducing misalignment easier than alignment, easier to poison than to make robust etc. Bias oft embedded in training data...

010

amatheac.bsky.social @amatheac.bsky.social · 16/09/2026

Three syllables in a relatively unfamiliar arrangement and it is rapid onset illiteracy with some people.

020

amatheac.bsky.social @amatheac.bsky.social · 16/09/2026

I am Jill's complete lack of surprise. Centrists are just cowardly fascists on average.

000

amatheac.bsky.social @amatheac.bsky.social · 16/09/2026

It is in their nature to be resistant to those priorities in the backdrop of a fascist status quo.

000

amatheac.bsky.social @amatheac.bsky.social · 16/09/2026

"Benjaun" , like she couldn't even bother to get your name right. FFS.

120

amatheac.bsky.social @amatheac.bsky.social · 15/09/2026

I have seen more evidence of a spine in literal invertebrates. For Fuck's Sake.

000

amatheac.bsky.social @amatheac.bsky.social · 15/09/2026

Also back then they actually made you look at an ad on your way to blocking someone , genius marketing strat if you ask me, since there was a *lot* of blocking involved.

000

amatheac.bsky.social @amatheac.bsky.social · 15/09/2026

Been years since I last used Grindr, but one of the more memorable experiences I had on there (amidst a deluge of unsolicited dick pics) was an unsolicited butthole pic. Just 'cause the meme was that we were all catgirls , it didn't follow that mimicking a feline greeting was appropriate!

100

amatheac.bsky.social @amatheac.bsky.social · 15/09/2026

Looks like Ed Sheeran lacks moral fortitude just as much as he lacks intrinsic resistance to sunburns.

000

amatheac.bsky.social @amatheac.bsky.social · 15/09/2026

Also remember the fuckery in Libya and her appalling "we came, we saw, he died" self-congratulatory gloating? Great fucking job destabilising Libya to the point overt slavery and slave markets - mostly trafficking enslaved Black people, returned at a scale unprecedented for some time.

0313

amatheac.bsky.social @amatheac.bsky.social · 14/09/2026

Dweebs like that conveniently forget that immigrants tend to be far more liberal than counterparts in countries of origin , and also that 1st/2nd gen immigrants tend to skew center-left/left. Later generations tend to mirror local population, generally. academic.oup.com/isq/article/...

In many regions of the world, liberal politics is on the retreat. This development is usually explained with reference to inherently political phenomena. We propose an alternative explanation, linking democratic backsliding to deep-reaching demographic change caused by mass emigration. We argue that because migrants tend to be more politically liberal, their departure, if quantitatively significant, can hurt liberal democracy. Empirically, we focus on Central and Eastern Europe (CEE). Since 2004, the region has lost about 9 percent of its population due to migration to Western Europe. Drawing on data from 430,000 individuals and a panel analysis, we show that CEE migrants systematically hold more liberal values than non-migrants and that their exit went along with a deterioration of democracy in their home countries. Further analyses show that the mechanism we describe generalizes to various other world regions. Mass emigration may pose a challenge to democratic development in migrant-sending countries around the globe.

020

amatheac.bsky.social @amatheac.bsky.social · 14/09/2026

Some of them might be the first documented examples of male-to-male transsexuals, lowkey.

010

amatheac.bsky.social @amatheac.bsky.social · 13/09/2026

"how many people have ever had experience x" and "what is the total burden of x" are two completely different questions. Asking the former when answering the latter in terms of approach is beyond incompetent.

000

amatheac.bsky.social @amatheac.bsky.social · 13/09/2026

With studies like the USTS we have the opportunity to generate a significant evidence base that more accurately represents the total burden of transphobia + make case for dual social/medical necessity of FFS etc , but 2 surveys in a row, the methodology appears to have squandered the chance.

100

amatheac.bsky.social @amatheac.bsky.social · 13/09/2026

A thing that chafes me re: sexual/intimate partner violence research is recording data as never vs ever or yes/no in the past year. Completely erases repeat victimisation. Also wrt trans people - we fail to record transition stage (pre / HRT / top surgery/ FFS etc) wrt shaping risk profiles.

100

amatheac.bsky.social @amatheac.bsky.social · 12/09/2026

They're ironically both part of the same thing - stigmatisation of transexual bodies. Medical gatekeeping seeks to morally mandate transexual bodies out of existence through denying care when it produces visibility. Anti-transitioners in trans spaces want to shame transexual bodies to the same end.

0140

amatheac.bsky.social @amatheac.bsky.social · 10/09/2026

I wonder whether Paul Verhoeven is saying to himself "I thought I made the satire in my Starship Troopers movies obvious enough to even the most terminally media-illiterate!" in disbelief.

000

amatheac.bsky.social @amatheac.bsky.social · 09/09/2026

Academia is kind of brutal fr. I just had a crazy little phase where I had to essentially go 50+ hrs for 8 weeks straight and do 100h over the last 10 finishing up a paper and a grant app. Glad to finally get to chill though.

000

amatheac.bsky.social @amatheac.bsky.social · 09/09/2026

TFW the 5090 in RTX5090 starts referring to the price lmfao.

020

amatheac.bsky.social @amatheac.bsky.social · 09/09/2026

Sorry to be that pedant - but DSR , DLSS and "DLSS"5 are three different things. DSR simply renders at higher res and then downscales, which effectively antialiases images better. DLSS 1-4 take lower res input frames and up-sample them. DLSS5 adds a bunch of stuff on top of frames (yassification).

110

amatheac.bsky.social @amatheac.bsky.social · 07/09/2026

Just going to point out one of these things is not really like the others, because there is far more buy in into right-wing bigotry from cishet white women than there is from the other marginalised groups mentioned. Cishet white women and cishet Latino men both played key roles in ushering in Trump.

010

amatheac.bsky.social @amatheac.bsky.social · 07/09/2026

I've interviewed for tenure track positions where getting an interview was easy but offers never followed. Experiences include having a cis female panel head call being trans "a lifestyle choice", being questioned about my own contributions to my co-first authored nature paper, etc.

0182

amatheac.bsky.social @amatheac.bsky.social · 02/09/2026

How appalling that there is finally a medical setting where doctors actually manage to shut the fuck up and listen to patients they usually tune out! (paraphrased subtext).

020

amatheac.bsky.social @amatheac.bsky.social · 02/09/2026

Doesn't have to be patented for gatekeeping - pharma grade E2/P4 is prescription-only & OTC products are very low dose (to treat hot flashes etc). You CAN enhance absorption of OTC E2 with scrotal admin etc (I DIY'd this way first 7 months) but labia minora are less absorptive in contrast.

110

amatheac.bsky.social @amatheac.bsky.social · 02/09/2026

What specific kind of cancer? For epithelial ovarian cancers HRT is fine (no evidence it is detrimental). For other ovarian cancer subtypes it isn't clear cut,since some have hormone receptor positivity that can fuel the growth of cancer cells therein. pmc.ncbi.nlm.nih.gov/articles/PMC...

pmc.ncbi.nlm.nih.gov

Hormone replacement therapy after surgery for epithelial ovarian cancer

Women who have undergone surgical treatment for epithelial ovarian cancer (EOC) may develop menopausal symptoms due to immediate loss of ovarian function following surgery and chemotherapy. Women may ...

000

amatheac.bsky.social @amatheac.bsky.social · 02/09/2026

Breathtaking lack of self awareness from the clickbait industrial complex. PS - I'd appreciate if someone could send me screenshots / copied text of the entire article. I am irritated enough to address any and all bullshit claims made in the article in some depth.

000

amatheac.bsky.social @amatheac.bsky.social · 02/09/2026

anti-ADD/autism rhetoric is also per se an instrument of transphobia given disproportionate coincidence with GD ; the first is invoked to prop up accusations of fabulism to discredit ADD trans people of being unreliable narrators of medical experience. The latter is invoked to minimise autonomy.

020

amatheac.bsky.social @amatheac.bsky.social · 07/08/2026

Cis women siding with cis men to uphold cisarchy. At this point I'm convinced no class of cis people is safe.

040