@C_Harwick i've been flummoxed by this question since reading the "Rented Virtue" Manidis essay. at the very least, contemporary free marketers have to pay regard to the Quaker insistence on fixed prices, given that they were the ones who invented capitalism.
An underrated observation in the latest ESM paper is that you shouldn't use the last layer of embeddings as features for downstream prediction tasks. (Most of the papers that supposedly debunked the value of PLMs for downstream predictions used the last layer. Oops!)
@sameQCU for me, this is now true, but it only became true recently due to coding agents. without agents, i didn't have much energy after work to grind on low-level coding details. with agents, not having enough projects is bad for productivity because i start scrolling instead
i do think there's a fair amount of low hanging fruit to inculcate into llms the principle of sufficient reason as a guide for research. eg when trying to improve a method after getting benchmark results, claude doesn't tend to form a theory to explain patterns in the benchmark, then propose improvements based on that theory; it tends to propose improvements that were already plausible prior to getting those results. but this doesn't seem so hard to fix.
This is a very important question even outside biology. I don't think current AI models would have looked at Uranusβs orbit and theorized the existence of Neptune, or looked at neural spike trains and theorized the Hodgkin-Huxley model.
New Blog: What's the point of theory in biology, especially in the age of machine learning?
I just published a series of letters by @NoahOlsman that start to get at this question, especially in the context of virtual cells: https://t.co/tcJLPgPs94
wokism: even though it's way better for the people who adhere to the frequent puffs, lower SES people don't adhere. so no one is allowed to get it because equity blah blah
capitalism: it's generic, so big pharma doesn't like it
scientism: it was discovered by a mad scientist who experimented on himself, then broadly adopted without RCTs. so evidence based medicine ideologues don't like it
@HououinTyouma@zeta_globin so mast cell activation is the fundamental driver of most asthma. personally, my asthma was best managed when i was able to get cromolyn sodium. but it's been abandoned for less effective, higher side effect drugs because of a woke capitalism scientism trifecta.
@HououinTyouma@zeta_globin cromolyn sodium is goated -- i'm pretty sure i've ranted about this before, but it's a travesty that inhaled versions are basically unavailable today, when they should be otc
the reason many non-catholics are excited by the ai encyclical is that they intuitively grasp the true role of the modern papacy: to act as a comforting caretaker of managed decline, as the church was replaced by the state. now, this expertise can be used to manage the replacement of humanity with agi.
Significant fractions of Magnifica Humanitas, the Pope's new encyclical on AI, appears AI-generated or AI-assisted.
Maybe Claude specifically, judging from the repeated references to 'genuinely."
Kinda interesting!
@EricRichards22 given that Boston isn't actually a big city but has bad traffic, having the airport centrally located isn't so bad imo. eg, i get why O'Hare is far from Chicago, but it's really idiotic when a place like Pittsburgh has its airport over an hour away.
@curious_vii@krishnanrohit charging per hour makes sense, but i don't think they'll be able to increase their hourly rates that much. the supply of consulting-ish labor has skyrocketed, so they're facing a deflationary shock that will counteract increased productivity.
TIL: "traditional attention in transformers reduces to the special case of maximum a posteriori (MAP) inference of values given queries,
assuming Gaussians for the likelihoods" https://t.co/ujJ2JR4BGv