Post B90gwjCYRhViWicP2G by ariadne@social.treehouse.systems
 (DIR) More posts by ariadne@social.treehouse.systems
 (DIR) Post #B90g4KcTcn4IeLnyeO by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       the problem with discussing the actual state of machine learning is that you get people who just show up in your mentions wanting to debate some strawman that you never discussed in the first placeit is exhausting
       
 (DIR) Post #B90gD4UOxwejEGCBJQ by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       one side, carrying economy-breaking capital pressure, spreads misinformation about the capabilities of their silly LLMs (which are silly)other side, out of rightful and just anger of the capital side, spreads misinformation based on strawman arguments that have nothing to do with what is being discussed
       
 (DIR) Post #B90gIXTOdGHOZJKfJI by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       incidentally: i don't have respect for either side
       
 (DIR) Post #B90gZlzZu6boWObRkO by cadey@pony.social
       0 likes, 0 repeats
       
       @ariadne and people will randomly assign you to either side without either hearing your opinion or displaying judgement
       
 (DIR) Post #B90gwjCYRhViWicP2G by ariadne@social.treehouse.systems
       1 likes, 0 repeats
       
       @cadey i think in your case, it's a little different, right?  because you are using LLMs to actively develop anubis.
       
 (DIR) Post #B90hS1kfWhVr2JV2Y4 by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       my interests are, at this point, essentially harm reduction:- debunking bullshit when i see it (such as yegge's post today)- pushing for more ethical research approaches (we do not need to build giant unscalable monoliths)- pushing for more clear definitions so that people are equipped with a sufficient understanding of what is actually going on for real
       
 (DIR) Post #B90iOL220AN3d0QS00 by wronglang@bayes.club
       0 likes, 0 repeats
       
       @ariadne Amen.... I was mad at machine learning self-sabotaging itself by focusing on MSE (or adjacent) as the metric to evaluate all algorithms on (almost) all problem domains long before the LLM era and I'm just annoyed at how predictably the field failed to manage the LLM transition.
       
 (DIR) Post #B90iX6Js4SHYbZwrh2 by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       in my opinion as someone who has exhaustively researched ML and has owned production ML workloads at adtech-scale in the past, i think a large part of the problem is that LLMs have enabled people to sound confident while not knowing a fucking thing about what they are talking about."look ma, i'm an AI engineer now" energy
       
 (DIR) Post #B90ilQLKQ958lFSZFY by ariadne@social.treehouse.systems
       0 likes, 0 repeats
       
       and let me tell you: just because you know some of the lingo, and read some of the papers, doesn't mean you know how to build scalable ML workloads that are production-quality.this LLM shit ain't it, chief.  it is a scalability nightmare.  MoE fixes some of it by making sharding easier (as demonstrated by colibri), but it doesn't cure the scalability problem of having a trillion+ parameter model.but at the same time, there is meaningful research happening.
       
 (DIR) Post #B90n3xnNBTNfXafnhw by deutrino@mstdn.io
       0 likes, 0 repeats
       
       @ariadne the drumbeat of in-group signalling about genAI on here is at an absolute fever pitch. it's expanded to ~20% of my feed and most of those posts are utterly redundant and not worth reading. my filter set is expanding by dozens of terms a day.maybe now is a good time for folks to pivot from loudly ranting about the state of things and proclaiming that only good people agree and bad people disagree, to discussing what they'll do about the situation as it is (and not some made-up story).
       
 (DIR) Post #B91OoC0yLUZhGYwQ08 by dvshkn@social.treehouse.systems
       0 likes, 0 repeats
       
       @ariadne news about machine learning that's explicitly not LLMs would be kind of refreshing, I supposed I could go get good at one of the arxiv front-ends to try and fetch it myself
       
 (DIR) Post #B92Hw3NvoFN4IURQLw by hazel@social.treehouse.systems
       0 likes, 0 repeats
       
       @ariadne you think this is a pain for you. try being the person who invented the idea of finding bugs with language models
       
 (DIR) Post #B93TkZXboypdA1Dfyi by ska@social.treehouse.systems
       0 likes, 0 repeats
       
       @ariadne It really is the problem with discussing anything on the Internet. You only notice it with discussions about the actual state of machine learning because it's close to a very hot topic, so a lot of people want to jump in.And yes, it is exhausting.