sarahconstantin's Shortform

sarahconstantin

https://roamresearch.com/#/app/srcpublic/page/10-11-2024

https://www.mindthefuture.info/p/why-im-not-a-bayesian [[Richard Ngo]] [[philosophy]] I think I agree with this, mostly.
- I wouldn't say "not a Bayesian" because there's nothing wrong with Bayes' Rule and I don't like the tribal connotations, but lbr, we don't literally use Bayes' rule very often and when we do it often reveals just how much our conclusions depend on problem framing and prior assumptions. A lot of complexity/ambiguity necessarily "lives" in the part of the problem that Bayes' rule doesn't touch. To be fair, I think "just turn the crank on Bayes' rule and it'll solve all problems" is a bit of a strawman -- nobody literally believes that, do they? -- but yeah, sure, happy to admit that most of the "hard part" of figuring things out is not the part where you can mechanically apply probability.
https://www.lesswrong.com/posts/YZvyQn2dAw4tL2xQY/rationalists-are-missing-a-core-piece-for-agent-like [[tailcalled]] this one is actually interesting and novel; i'm not sure what to make of it. maybe literal physics, with like "forces", matters and needs to be treated differently than just a particular pattern of information that you could rederive statistically from sensory data? I kind of hate it but unlike tailcalled I don't know much about physics-based computational models...[[philosophy]]
https://alignbio.org/ [[biology]] [[automation]] datasets generated by the Emerald Cloud Lab! [[Erika DeBenedectis]] project. Seems cool!
https://www.sciencedirect.com/science/article/abs/pii/S0306453015009014?via%3Dihub [[psychology]] the forced swim test is a bad measure of depression.
- when a mouse trapped in water stops struggling, that is not "despair" or "learned helplessness." these are anthropomorphisms. the mouse is in fact helpless, by design; struggling cannot save it; immobility is adaptive.
  - in fact, mice become immobile faster when they have more experience with the test. they learn that struggling is not useful and they retain that knowledge.
- also, a mouse in an acute stress situation is not at all like a human's clinical depression, which develops gradually and persists chronically.
- https://www.sciencedirect.com/science/article/abs/pii/S1359644621003615?via%3Dihub the forced swim test also doesn't predict clinical efficacy of antidepressants well. (admittedly this study was funded by PETA, which thinks the FST is cruel to mice)
https://en.wikipedia.org/wiki/Copy_Exactly! [[semiconductors]] the Wiki doesn't mention that Copy Exactly was famously a failure. even when you try to document procedures perfectly and replicate them on the other side of the world, at unprecedented precision, it is really really hard to get the same results.
https://neuroscience.stanford.edu/research/funded-research/optimization-african-killifish-platform-rapid-drug-screening-aggregate [[biology]] you know what's cool? building experimentation platforms for novel model organisms. Killifish are the shortest-lived vertebrate -- which is great if you want to study aging. they live in weird oxygen-poor freshwater zones that are hard to replicate in the lab. figuring out how to raise them in captivity and standardize experiments on them is the kind of unsung, underfunded accomplishment we need to celebrate and expand WAY more.
https://www.nature.com/articles/513481a [[biology]] [[drug discovery]] ever heard of curcumin doing something for your health? resveratrol? EGCG? those are all natural compounds that light up a drug screen like a Christmas tree because they react with EVERYTHING. they are not going to work on your disease in real life.
- they're called PAINs, pan-assay interference compounds, and if you're not a chemist (or don't consult one) your drug screen is probably full of 'em. false positives on academic drug screens (Big Pharma usually knows better) are a scourge. https://en.wikipedia.org/wiki/Pan-assay_interference_compounds
- sadly, while they make automated PAINs alerts, they don't work for shit. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5411023/ sorry, shut-ins and cheapskates; you might have to talk to an actual chemist.
https://en.wikipedia.org/wiki/Fetal_bovine_serum [[biotech]] this cell culture medium is just...cow juice. it is not consistent batch to batch. this is a big problem.
https://www.nature.com/articles/s42255-021-00372-0 [[biology]] mice housed at "room temperature" are too cold for their health; they are more disease-prone, which calls into question a lot of experimental results.
https://calteches.library.caltech.edu/51/2/CargoCult.htm [[science]] the famous [[Richard Feynman]] "Cargo cult science" essay is about flawed experimental methods!
- if your rat can smell the location of the cheese in the maze all along, then your maze isn't testing learning.
- errybody want to test rats in mazes, ain't nobody want to test this janky-ass maze!
https://fastgrants.org/ [[metascience]] [[COVID-19]] this was cool, we should bring it back for other stuff
https://erikaaldendeb.substack.com/cp/147525831 [[biotech]] engineering biomanufacturing microbes for surviving on Mars?!
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8278038/ [[prediction markets]] DARPA tried to use prediction markets to predict the success of projects. it didn't work! they couldn't get enough participants.
https://www.citationfuture.com/ [[prediction markets]] these guys do prediction markets on science
https://jamesclaims.substack.com/p/how-should-we-fund-scientific-error [[metascience]] [[James Heathers]] has a proposal for a science error detection (fraud, bad research, etc) nonprofit. We should fund him to do it!!
https://en.wikipedia.org/wiki/Elisabeth_Bik [[metascience]] [[Elizabeth Bik]] is the queen of research fraud detection. pay her plz.
https://substack.com/home/post/p-149791027 [[archaeology]] it was once thought that Gobekli Tepe was a "festival city" or religious sanctuary, where people visited but didn't live, because there wasn't a water source. Now, they've found something that looks like water cisterns, and they suspect people did live there.
- I don't like the framing of "hunter-gatherer" = "nomadic" in this post.
  - We keep pushing the date of agriculture farther back in time. We keep discovering that "hunter-gatherers" picking plants in "wild" forests are actually doing some degree of forest management, planting seeds, or pulling undesirable weeds. Arguably there isn't a hard-and-fast distinction between "gathering" and "gardening". (Grain agriculture where you use a plow and completely clear a field for planting your crop is qualitatively different from the kind of kitchen-garden-like horticulture that can be done with hand tools and without clearing forests. My bet is that all so-called hunter-gatherers did some degree of horticulture until proven otherwise, excepting eg arctic environments)
  - what the water actually suggests is that people lived at Gobekli Tepe for at least part of the year. it doesn't say what they were eating.

everybody want to test rats in mazes, ain't nobody want to test this janky-ass maze!

One of the interesting things I found when I finally tracked down the source is that one of the improved mazes before that was a 3D maze where mice had to choose vertically, keeping them in the same position horizontally, because otherwise they apparently were hearing some sort of subtle sound whose volume/direction let them gauge their position and memorize the choice. So Hunter created a stack of T-junctions, so each time they were another foot upwards/downwards, but a... (read more)

sarahconstantin's Shortform

8

sarahconstantin's Shortform

8