YoungAchamian
No bio...
User ID: 680
YoungAchamian
3hr ago
We seem to have a different definition of what constitutes "ML Research". I'd break it down into two forms: Basic Research and Applied Research. Basic is probably not the precise word because a lot of core research is non-basic, but core is also an imprecise word, as is making a boundary around theoretical.
But Applied Research is pretty straight forward. It is the application of ML theory and algorithms/models to real-world practical problems. The "Basic" Research is generally more on developing the ML theory of what can work or is possible. You seem to think Applied Research is not actual "ML Research". I'm not sure the ML community agrees with you because there are prevalent conferences like CVPR or the NLP one I am blanking on. These are considered ML conferences, focused on a particular practical field. Industry research is almost always Applied, not all of us have the luxury of working on grants, business want returns and the research is around applying ML theory to real-problems. Like the Cuffless BP Nature paper. I think your definition is overly purity focused, though I imagine our tension is one as old as time between Academic PhDs and Industry Researchers.
The last two are definitely the "core/theoretical/basic" side of research because they aren't actually applying it to real problems. One's just a theory on Causal Modeling al la Pearl or Schölkopf. The pipeline is that someone like me takes these more theoretical models and implements them in the real-world.
Grassmann manifolds are outside of standard ML math, but the explanation in 2.2 was easy to follow.
Maybe I suck at math (a real possibility) or maybe you are just good at math (also a possibility) I still am very shaking on what a Grassmann manifold is. I don't think the paper is earth shattering in itself. I've seen several papers about kernelizing attention, or linearizing it, or anything to make it non-quadratic.
Again, this doesn't seem very mathy to me
I don't think this one is mathy, but it is arcane on the applications of meta-learning as bayesian priors to allow a model to generalize across out of distribution problems during inference time. Claiming it can do zero-shot inference on unrelated tasks because it learns how to formulate problems as an approximation of bayesian inference in a practical amount of time is a wild idea. It's making a very complicated claim that takes a long time to wrap your head around.
Having to look up 3 references to read and understand a paper seems absolutely reasonable to me.
Unfortunately this is a constraint in industry, I have a job, there is work to get done. spending 8+ hours to digest a theory paper is a large impact on my time. Even if it leads to something useful.
I am a ML researcher, in Industry without a PhD. The papers are absolutely for me. (And if they aren't then thats a major clique/circle-jerking issue, as I'm the one actually trying to apply what is being done)
https://www.nature.com/articles/s41598-025-07087-2 This paper I recently tried to replicate for research on IoT cuffless BP, it absolutely fails to replicate. Not only that, but it also suffers from massive subject leakage on how it splits the data. It's pretty much overfit with a 75% overlap between signals and then it shuffles those between train and val. Even copying it's splitting approach I failed to get more than a MAE SBP of 6.07 and DBP of 4.3. Paper claims sub 2.0 for both.
Then there's this: https://arxiv.org/pdf/2512.19428. Maybe you know Grassmann flows and manifolds but I definitely did not learn this naturally. I pretty much need a background tutorial on this.
I actually enjoyed this paper's concept: https://arxiv.org/pdf/2602.14972 But needing to read 2-3 additional papers, one of which was super mathy proving out the intuition was a lot of work. It still takes me a bit to conceptualize this because it is DEEP in the bayesian world.
Maybe you are in a different subfield than I am, but I have consistently failed to replicate paper results for the occasional paper for the last 4-5 years. It happens, it's a thing. If I say that to other industry researchers they pretty much agree. One of the reasons we think poorly of academics.
YoungAchamian
1d ago
It's extremely bad in ML literature, where pressure to publish and get your citation + h-index up means that people publish all kinds of non-replicate-able junk. Nobody wants another incremental advancement paper so every paper is revolutionary. I cannot tell you the number of times I've taken a paper that looked interested, either used their code or re-implemented it approximately, used their datasets and gotten far worse results. The new trick, or really old trick making a resurgence, is to include all kinds of arcane math in the paper and not provide any code so its impossible to replicate it without a math PhD in that area.
YoungAchamian
1d ago
The evo-psych explanation works in this case for me. I have theorized additional factors related to her personality that I think strongly impact why this occurs in their relationship.
The amusing thing to me is that these two are highly progressive (the husband is pretty much a champagne socialist) but lead a very trad lifestyle. She likes to spin it as being poly, but she's clearly just a cuck.
YoungAchamian
1d ago
I have a friend(The wife) who seems to have this predilection. Though it's not cheating because it is explicitly open on his side. But she seems to get off on knowing that other women find her husband hot and that he's high value enough to still get laid on the reg.
I personally don't think he's really the type to actually do it, and personally don't think he's that much of a looker besides being tall, but a girls gotta have her fantasies and as far I know he doesn't disabuse her of the notion.
YoungAchamian
2d ago
I was according to the quiz 46% Autistic and 56% German. I was also formally diagnosed as a child around like 8(??), though at the time my diagnosis was Aspergers. Which has been rolled into a general "on the spectrum". I'm pretty high functioning, and my special interests have help me mostly pass.
My girlfriend got 30/60 German/Autistic on the quiz and only really passes because she's quiet af so people don't notice or clock her.
YoungAchamian
2d ago
What do you think of AlphaFold
Is not an LLM, its a Diffusion model. If you are going to call someone out bad AI takes then I'd recommend you wrap your head around the AI vs LLM distinction.
AI is still completely useless at real scientific research and reasoning.
It's not doing reasoning via any sort scientific deduction. There is a whole subfield of AI called causal discovery around trying to get models to learn via a causal learner. If you want substance I can fetch you some papers, there are plenty. None of them are LLM papers.
YoungAchamian
4d ago
Are trivia organizers compensated? It's not clear to me if they are or not. It's definitely not a full time gig. But yes its a lot of work, welcome to why third spaces don't exist.
YoungAchamian
4d ago
I mean this is needlessly picky, its like the one-drop rule for individualism. If you are even a drop of individualist, you are not a true collectivist. By that definition Communists are not collectivists, neither are progressives, pretty nobody but an actual hive-mind is a collectivist. A simpler and more realistic boundary is that you you talk about apply policies actions to a collective, you are a collectivist.
YoungAchamian
4d ago
theories of disparate impact which are very collectivist.
I don't disagree that disparate impact theories are collectivist but fighting a specific collectivism doesn't make you not a collectivist. The easy answer is you just want the collective to favor your theories instead of others.
"reshape society"
Idk, change policies, laws, culture so that certain theories are now fundamental to the fabric. If you think African Americans really are genetically less intelligent, and more prone to violence, do you really mean to tell me that the response to accepting that is: "Well thats neat but nothing should be done about it" or are there policy actions that people want to put in place in order to curtain all of that. There are clearly dissident right voices that want to use HDB for policy actions, the autistic folks just want it to be "this is truth, we should stop hiding it" but in a way they are being naive or useful idiots to the class of people who actually want to use those theories to change society. It's like Autistic Marxists being naive about what the hard core revolutionaries actually want to do to force a communist society.
YoungAchamian
4d ago
Its socializing without putting effort into the space. If the point of the third space is just a place that you can show up, and requires nothing from you, no effort, just a lazy bum selfishly taking and providing nothing, then its no wonder all the spaces died. People justifying being a selfish taker socially is a major reason there are no third spaces.
YoungAchamian
4d ago
To be clear I am not an AI bear of @ChickenOverlord's persuasion, I am an AI bull, but very cynical on the marketing hype. I think his opinion that they are useless for coding does not align with my opinion. But I do think the original goalpost that Mythos is not some super intelligence, first step to the singularity, type model has not been moved. Mythos is a SOTA tool like other SOTA tools.
The amount of steering it needs is still totally speculative.
Absolutely, but reasoning by abduction points it to being worse than the marketing would suggest.
I understand the distinction might be too subtle to notice.
No need to get snarky. Pretend for a moment that I am very good at noticing. Consider that the modal alt-rightist wants all the progressive gibs but for white people, thats pretty collectivist by any definition
Anytime a political creed starts with "and this [group of people by an attribute] needs to have xyz done to them or given to them", then boom you are in the collectivist category. Pretty easy boundary line.
EDIT: I remembered a couple more examples: Dread Jim is part of the right, he and similar birds of the feather are pro-men/anti-women collectivists. They wants spoils and policy benefits that benefit men as a class and hurt women as a class. We have several people here on the motte that are in this camp. This is right-wing. Classic collectivism.
We have our resident joo-posters/neo-nazis, again would be classified as rightwing. They are clearly anti-jewish/Pro-white collectivists. They in particularly want gibs towards white people much like the progs do.
YoungAchamian
4d ago
I'll make a note to ping you specifically when it comes up again. HBD is boring to me, so there's no friction to make me remember this incidences and the search bar is functionally useless.
EDIT: I think we had a recent, past year?? flameout by some black dude with examples about all the nasty shit HBDers say about black people with links.
YoungAchamian
4d ago
AI researcher was a skilled profession 15 years ago
Lmao, AI research was an extremely niche profession 15 years ago. AlexNet came out in 2012 and pioneered deep learning which lead to transformers/LLMs. The only people who were AI researchers were college professors and the occasional industry researcher. You had a number of folks doing data science/ML engineering but it was pretty much just a sub-domain of statistics as thats what boosting/SVMs/Random Forests are.
YoungAchamian
4d ago
nothing about collectivism
We must witnessing very different applications if you think the average HBD poster is making comments about African-Americans being more violent and lower IQ on an individual level. And not by definition on a collective level. There's a fig leaf towards it being an distribution and obviously not every individual. Followed up by here's my 10 step plan to reshape society so that AAs collectively have reduced social impact, freedom, rights, and political power.
HBD is a belief about an is, not an ought
Only in the most theoretical autistic form. If the belief is that certain populations underperform along ethnic lines and have increases in certain undesirable traits. The follow on is almost always policy actions to reshape society around that theory. That's an "ought" not an "is"
YoungAchamian
4d ago
I mean maybe if Republicans gave up on Abortion, Gay Rights/Marriage/LGB, and Donald Trump, then yeah. But I wouldn't hold my breath. And if you think marriage and children is a conservative pipeline, then I have a bridge to sell you in the Sahara. I know more Prog-Commies with kids or trying for kids than I know conservative men with kids. Turns out that conservative male politics outside of a religious community makes you unmarriageable to most normie women.
default preferences of college aged women
Square this fact with the Republican modal beliefs on women and tell me how they converge in the slightest.
YoungAchamian
4d ago
In my conception of effort, the effort required is not the people showing up to trivia, its the organizer. The one who gets the questions every week, contacts restaurants and bars for hosting, markets the event, etc. Regulars are great and you can't have a reliable third space without them, but if that organizer did not organize it, there would be no event. And if enough organizers in the local area don't organize it, there is no local trivia community.
YoungAchamian
4d ago
Do you not feel like the goal posts are shifting here?
The original claim Anthropic made was that Mythos could do all of this independently. That it didn't need a highly experienced security researcher guiding it. In fact that the reason it was so dangerous is because any lay-person could use Mythos to "hack the planet" It's not goal post shifting to point out, no Mythos is just a SOTA tool and like most SOTA AI tools it works better by having an experienced human guiding it on what to code, look for, design the system etc. The AI ecosystem is very hype oriented, people claim far more than what is realistically delivered.
YoungAchamian
4d ago
A fair assumption, my b for not being more clear.
Sorta, if your goal is to just show up at the bowling alley or bar and not provide anything, the "socially acceptable loitering" I don't think you can complain when the space shuts down because it ran out of business by having too many free loaders. You don't need to be a try-hard at drinking or bowling but you should pay for a lane, buy a few drinks to nurse, and contribute to the community. In my specific case, it's "nominally" a board game hobbyist group but its more of a social space. We do a lot of non-board game things, social events, dinners, etc. But there are people that contribute to the community/space in the form of hosting/planning events, bringing supplies, food, and there are people that just show up, give no effort other than their presence. A community/space cannot survive with a too large majority of the latter.
I'd struggle to think of any hobbyist group that is not some form of third space. The simple law of reality is that in order for something to exist, someone needs to make it exist. And that requires effort. Legislating a third space into existence does nothing. Someone needs to actually go organize the community garden, the local pub, the dance hall. And if they get hit by a bus, and the third space falls apart? Well then someone else needs to step up and do the work.
Yeah I wish... I might just be a midwit but I found ECE to be particularly difficult and laborious, and occasionally very abstract. I deliberately switched out of the CompE side of the major because the PCB/ASIC design courses were required and had horrible(deserved) reputations. A major component of it was that the department prided itself on "no grade inflation" and having lots of smart research professors. This translates into most courses being graded on actual curves in that 50% of the class fails regardless of absolute score, and the lectures being completely pointless to attend (but losing points for non-attendance in a way that could only hurt you). Most learning was done in TA study halls, Professor office hours, and at home. It was essentially learn this by yourself and we'll grade you, oh and learn it better than your classmates.
I remember the average on an intro linear circuits exam (the Thevenin's equivalence topic area) being something like a 28% I had been chatting with the professor and she pretty much admitted they had went a bit overzealous with that one and actually felt bad.
It was a slog, and still to this day I feel like I suck at math.
YoungAchamian
4d ago
Having interacted with actual sorority girls in the last decade, even in my flyover midwest state school, yeah no. I wouldn't call them dyed in the wool progressives but thinking they are reliable GOP votes is equally fantastical. Obviously we didn't do a lot of political talking but my recollection is that most of them had normie feminist politics, with a splash of family oriented-ness and occasionally some cultural christianity
They want third spaces to exist, but they don't want to start them or put the required effort into them to make them flourish. Most people are just free riders that want the benefits. I run a local board game meetup and trying to get people to do anything more than show up (and even then) is very hard.
Ok I can grant that the conference communities are distinct. But I do think there is still some difference between those two papers and papers like these: CATs and DAGs: https://arxiv.org/abs/2410.14485, Distributed Alignment Search: https://arxiv.org/abs/2303.02536. I had a solid idea of what these papers were doing in ~10 minutes of reading them. But it took me several hours each to understand the Prior-Fitted Networks, then the math paper behind it, then the Causal Prior-Fitted Networks and lastly the Causal Foundation Models with partial information, at a level that I could apply or implement them. If you are saying that's because I am >20 textbooks and >1k papers behind on the background understanding, then sure I can't argue with that. But I could implement CATs and DAGs and DAS within the hour of reading them.
If I am not the target audience for any of them, how come some are more approachable from a practical side and others are not? I would expect truly being an outsider to have a roughly even level of penetration on the topics, not a wildly disparate one. To me that is indicative of a level of quality, skill on the author, or even writing to a wider audience (pretty much a skill) instead of writing to the clique (poor intent).
"Too" is a load bearing word that I don't think I have used. My complaint is some of these are fairly arcane, use complicated math of obfuscate a straightforward idea, or are deliberately being written in a way to gatekeep.
Some industry labs are essentially better paid academic labs, and attract prestigious academics. Google, Deepmind, FAIR, Microsoft, and plenty of small startups targeting PhD students. OpenAI started out that way. Other industry labs/research companies are far more implementation focused, in fact I'd say the majority are the latter. You'll get PhDs but also ML Engineers (like me) who transitioned as time/interest went on. And yeah we publish in the CVPRs and ACLs. ImageNet was a CVPR publication, AlexNet was a Nature publication. If you'd like to say those aren't "ML" then I'd say you definitely have a purity issue, if you want to grant them access to the vaulted halls, then its farcical to say that CVPR is not part of the ML community (doesn't need to be exclusively, I'd grant that the CV community and the ML community have not always had overlap)
More options
Context Copy link