there are helicopters twice a day overhead. i assume the noise sends a message. every time, a single neuron is diverted into ensuring the blades of fascism recede into the distance. twice now, i have envisioned an alternative, which quickly requires active thought aversion.
Post
Replies:
0
i'm sure i can assemble a ragtag group of hackers to build an insane microkernel
we are cancer researchers, and we give thanks to our patients. we tell them: i do not expect to cure cancer in my lifetime. we show them: i will fail at my single goal.
leukemia induces fear in my head that won't go away. it's not fear: i see it in their blood vessels when i wake up, not mine. i can't solve it yet.
matrices are different when you know what each cell means. every chapter and verse of every protocol. they winced but they were so brave for row 7, days 3-13. someone trusted you when they asked if this data really helps and you said "yes"
i've learned a strange and unsettling truth: biologists are being deskilled out of statistical application and data collection. biology journals don't want new methods. bioinformatics journals are peer reviewed by people who think they're smarter than cancer researchers and get scared if you ask them about a number they published. what the hell kind of scientist fears their own numbers?
@hipsterelectron maybe we need new paper
(i wonder if university admins like having fuckboy credit stealers in charge/tenured/etc because they can be manipulated. doubly so for data fabricators. literally a win-win for admin: data fabrication lets you do hot science at speeds no real scientist can match (without my help). if they ever rock the boat, leak their lies, wash your hands, repeat with new hire)
cc @inquiline i had never considered the self-reinforcing potential of the two-party system of uni admin vs fuckboy. maybe this is obvious and well-known? but it took me until just now to overcome my assumption that "people who lie are liabilities" (i admit i perhaps retain a Romantic idealization of academia)
i think there are structural incentives i can speculate on:
- R1 universities have lots of government funding. that's because taxpayers want to cure cancer. god i bet i could get some taxpayers real mad about this
- a 3-minute video walking through the software they are required to use to get published would result in lawsuits, if the government did that sort of thing.
- "lawsuits?" government grants have a famously lengthy (but standard) set of requirements, which used to be "do not violate the civil rights act" but also relate to concerns regarding fraud. particularly if the t-SNE fabricator machine were ever used in grant applications, progress reports, and other summaries, eventually if it becomes knowingly false someone (uni admin ideally, but they'll probably bounce it back)
- ok so software that so obviously and evidently does not work by design is not necessarily illegal but this shit is why i hate software corps. the expensive software, that cannot be reproduced, that is >30x slower, that actively functions by impeding the process of science, that is required by anonymous reviewers
i am slowly convincing myself that:
- anonymous reviewers are on the take
- uni admins are on the take
unfortunately "lying about numbers for money" is generally not considered harmful unless people directly die from it i think. also, both of these seem obvious now (anonymous reviewers are obviously not all the same person, that would be adjective). i bet a real uni admin would sneer at me for being surprised at that
you can't own the data, you can't parameterize their charts, they literally only support obsolete and discredited tools like t-SNE which literally just falsifies data (i shit you not. dimensionality reduction is fake but this one also adds noise cause it looks sciencey. first time i ever found a citation loop for a quantitative claim relied upon everywhere to do science. it would not be the last time)
if you didn't publish in wet lab biology in 2018 maybe the LLM assault on science is confusing to you but i did a paper in one of the hottest labs in the world with plenty of pub experience and they spent two years after i left getting it published because the paper is obviously groundbreaking in three distinct ways. https://pubmed.ncbi.nlm.nih.gov/30413431/
people whose actual job is publishing papers said it mystified them. that itself is a data point indicating that falsification and corpo deskilling is not just artificial but recent and accelerating.
this was all before twitter inc too lmao
because academic journals hear [read in kpop singer voice]:
hacker and doctor team up to take on cell differentiation and wrote their own analysis stack
and think [very evil hot queer villain voice]:
that sucks, nobody wants to hear that the expensive software that sucks is also wrong. you boy, throw some more copies of nature into the undergraduate dogfighting pit. add some sauce
[she's queer coded bc she will be joining our team later after we convince her that we are the one lab that doesn't lie. she will propose that we simply lie instead but we then explain that people will die and she gets all pouty but in the final battle she explains how we showed her how to hope and love. and she's also a hotshot biologist and we build mech suits together. platonically]
i don't ever again want to read a paper taking biopsies from unnamed dead people to chuck into the "literally data fabrication, this used to be illegal" machine. i'm removing a variable sized portion of your flesh for each dimension it reduces and pulling out a hair each time it adds noise to look more sciencey
@hipsterelectron jesus fuck what??
@davidgerard also "t-SNE fabricates data and is relied upon near-exclusively for single-cell protein analysis, both before expert input (gating) and very frequently for published figures" was both:
- true for many years
- mostly still true
- specifically profited corps, who know damn well what they're doing and didn't offer any alternatives
- uni admin hits ctrl-f for t-SNE, sends one-line email "Yes!", immediately gets blackout drunk
- fact-shaped data
- unlike proprietary LLMs, is math (can be disproven): like proprietary LLMs, is statistics (slur)
@davidgerard this is not a pitch because this is worth a blog post on MY site where i will ensure the reader at least is able to understand the abject kind of fear and nihilism i had about scientific truth........
where the punchline is:
(1) wow scientists are laborers!
(2) wow i didn't realize that assigning default white man from stanford the positive qualities of real scientists could harm others.....science was my comfort zone safe space!!
(3) there are no adults in the room and haven't been for a decade (in cancer bio, peer review is actively garbage, and other fields like physics have their own flaws, cryptography is unserious)
(4) bioinformatics is colonialism (reading ALL of fanon before i EVER put my name to words like that in public lol). i regularly benefitted from bioinformatics research funding White Guy Division and they just let you do whatever. these are people who refuse to read the works they cite. i would simply kms
@davidgerard oh sorry. yes this would be a good story for your column, if i was referring to a specific paper. unfortunately, i am combining multiple things together here:
- the datasets we used were single-cell cytof arrays from biopsies and blood draws. the names were redacted from me and i don't breach PII ever but i walked into a hospital (VUMC) which is a massive hub for cancer research. dr. irish is a really swell guy.
- when dr. greenplate describes my amazing R code which was written to her precise specifications and covered the whole analysis pipeline (this means she can hit run and get real fucking numbers from HER OWN statistical metrics, and immediately fuck around on her own) i feel like i fought an evil god and won but several years later when i read this i just feel how fucking disappointed this tech has ALWAYS been for her THE CANCER EXPERT.
- hence deskilling mention above—literally i was not an i/o genious at all then, i just wrote R that did the thing. her words: "massively increased scale" and especially "change over time". yeah the scientist who taught me how to evaluate statistics repeatedly calls me a fucking single-cell time lord
@davidgerard to me, it is excessively violent that people (literally kids) entrusted us as scientists to literally extract their flesh (dr. greenplate handled samples but idk if she directly interacted with them). to me the specifically LLM decontextualization is the act of violence here, cartoonishly evil. might as well dump these kids into a mass grave
@davidgerard maybe this will make it more clear why i'm making these very strong image-based analogies https://github.com/cosmicexplorer/comparisort
as literally one researcher who was literally looking for an excuse to implement mergesort (my algs prof liked it), and found "inconsistent ordering of proteins across studies" was literally my hole made for me
note that the sorting is very distinct from parsing here. the goal is not to parse and normalize, bc it's not a formal model. it's intended to represent a simple ordering structure simply
nobody has ever gone so far as to want to do look more like me when the vibe is "CANCER SCIENTIST SIDEQUEST! [Y/n]". nobody in any tech company will ever understand that users are not simply "more" or "less" expert. if the user is doing cancer research it's worth your time to sit the fuck down
i want to make people feel the rippling disappointment dr. greenplate describes with Literally All Scientific Software Ever. she then has a few lines about biopsies in the paper. i'm insane but she is not fucking around here.
literally all of bioinformatics is people who think they're me and will never be me because they will never listen to a woman tell them what to do. these people keep getting so much fucking money
@davidgerard radicalizing moment when my friend sam (nice guy) was trying to do real research w dan fabbri (useless fuckboy, bad lectures) was like "yeah a random forest keeps getting significantly improved accuracy over any neural net". he was cross-validating and shit. the data was afaict normal. fabbri just said "huh" and of course the work died instead of publishing a negative result.
the "huh" was a secondary source (sam relaying the "huh" to me) but i could absolutely tell he felt hurt and sad and upset and that fabbri was not a scientist he just gets paid to act like one.
that's also why i went out of the cs department and signed up for a med school class, bc i recognized the school actively does not want computer scientists thinking numbers mean things
@davidgerard so of course, i have decided that numbers mean biopsies of leukemia patients [literally true] and i swore to avenge their memory [every time i defeat a fuckboy one of their ghost voices says "thank you...." and gives me an additional health segment]
@davidgerard zero fuckboys so far. with the zstd length extension checksum collision proof of concept i will have nailed yann collett pretty good but i definitely need more mid-tier enemies to level up my cryptography stat.
there are two distinct cryptographer fuckboys and the one person who has actually read the paper i hate agrees that the paper is bogus (signal SPQR)
sorry for the false alarm and thx for checking in ^_^ !
if communion is jesus blood and flesh i guess that makes everyone who entrusted me to read their entrails was a legion of jesuses and i have like 17 free sins now sick
illegal black market prison economy of sin forgiveness vouchers that jesus and st. peter actually accept and vouch for. wardens in shambles. cigarette company CEO flees the country
i want to know academic journals that respect the contract we signed in blood with patients. i want to know academic journals who understand all the incentives against publishing cross-domain collaborative methods, who think hacker resistance is when computers are twisted to serve scientists—who, yes, are the underdogs now (big mistake)
hey isn't it so crazy how i keep finding tons of data falsification in completely separate areas of research? it does kinda sound like lots of money revolves around lies and the continuing construction of legitimacy. i mention legitimacy because i really do need people to say "google and microsoft are wartime assets of the us security state" whenever anyone legitimizes any form of collaborative work with or around them
read the BDS list for a thorough and profound set of receipts. i'm not here with receipts i think people should vilify and delegitimize microsoft and google open source. also if they touch the bytes you publish to your users you are exposing your users to an attack surface you can't reproduce (bc of the chainguard DRM scheme which is just a service that prints a checkbox you can't reproduce without the DRM)
(if you work at google or ms and you can read this please don't take it personally and also just make them pay you more if they're gonna make you accept moral injury)
