sarahconstantin's Shortform

sarahconstantin

links 12/13/2024:

https://arxiv.org/pdf/2407.00695 Minimo, an RL agent for jointly learning both conjectures and proofs in Peano from "intrinsic motivation"

what is "intrinsic motivation" in RL?
- https://arxiv.org/pdf/2203.02298 intrinsic motivation mechanisms include:
  - reward shaping, i.e. comparing the expected value of two possible states, so that the agent gets an incremental "reward" when it moves to a state with higher expected value
  - rewards based on novelty rather than expected success, such as assigning more reward to visiting novel states, or assigning more reward to states with high prediction error relative to the agent's model of the world
https://github.com/p-doom/gc-minimo gc-Minimo, the "goal-conditional" version that involves subgoals
https://pdoom.org/ AI organization, research aimed at AGI; young, educated European team, they seem smart (to my unsophisticated eye) and idealistic (they want to share/open-source as much as possible, in contrast to secretive for-profit AI labs)
https://news.mit.edu/2024/noninvasive-imaging-method-can-penetrate-deeper-living-tissue-1211 new non-invasive laser imaging technique; label free; 700 nm deep.
- aka, not useful for subcutaneous imaging in living mammals, but possibly quite useful for non-destructive imaging of organoids (mentioned in the article) or maybe invertebrates, embryoids, other small living things;
- maybe also nondestructive imaging of surface cells in live mammals:
  - skin
  - eyes
  - surgically exposed tissues
    - when you're operating on a tumor, it's important to make sure you have clean margins; would tumor cells look different under this sort of "metabolic" imaging?
https://xenaproject.wordpress.com/2024/12/11/fermats-last-theorem-how-its-going/ ongoing project to translate a proof of Fermat's Last Theorem into Lean.
- https://xenaproject.wordpress.com/what-is-the-xena-project/ the Xena Project is a project to get undergraduate math majors to formalize things in Lean.
  - "One could imagine things like formally verified course notes, which would later turn into some searchable database, and then to a tool which attempts example sheet questions by applying theorems from the course".
  - "No available system currently has all of an undergraduate pure mathematics degree, so undergraduates can even contribute to research projects. Over ten Imperial maths undergraduates have contributed to Lean’s maths library, and there are plenty of students at other universities in the UK and beyond who have also got involved."
https://reactormag.com/the-vampire-p-h-lee/ eerie, touching short story: what if, in early-2010's Tumblr, there were active vampire and werewolf communities?
https://www.za-zu.com/blog/playbook how to cold-email at scale. apparently if you just send a bajillion emails from one account it can get marked as spam; there are methods to circumvent this.
https://en.m.wikipedia.org/wiki/Kray_twins celebrity-esque 1960's British gangsters.
https://www.nature.com/articles/s41591-024-03306-x
- today in What Can't The Hypothalamus Do: stimulate the lateral hypothalamus and you get improved walking in recovery from spinal cord injury in mice, rats, and 2 humans.
  - appears to be specific to Vglut2 neurons (as shown by optogenetics)
  - got the patients to be able to climb stairs and walk 50 m, when they couldn't before, after 3 months of rehab (they had both had their spinal injury for many years prior without being able to walk/climb).
  - you can see from the emg data that both patients have way more leg muscle activation when trying to walk or raise their knees from a lying position when the DBS is on vs off
- how crazy is this? the standard lists of things the lateral hypothalamus does don't include motor function. mostly it's autonomic stuff, arousal, hunger, and motivation/mood.
  - otoh it does directly innervate the motor cortex, spinal cord, cerebellum, etc
  - https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2021.639313/full and there's some evidence that stimulating orexin neurons in the lateral hypothalamus induces movement (in mice) and hypothalamic orexin neurons are necessary for motor adaptation to sensory feedback https://www.jneurosci.org/content/42/32/6243
https://www.cognition.ai/blog/devin-generally-available this worries me from a mundane security point of view, though maybe I'm excessively paranoid; do you really want an AI agent autonomously mucking about in your code repo and pushing changes? I've heard the argument that this doesn't really introduce more risk than a new junior developer (who might likewise be error-prone or even a crook) but my mind is not at ease.
https://ideaharbor.xyz/ a cute site where people can post project ideas. some of them are not, y'know, possible. "Batteries that can store the internet in them for when your connection goes down."

sarahconstantin2y132

6Steven Byrnes2y

I’ve long had a tentative rule-of-thumb that: * medial hypothalamus neuron groups are mostly “tracking a state variable”; * lateral hypothalamus neuron groups are mostly “turning on a behavior” (especially a “consummatory behavior”). (…apart from the mammillary areas way at the posterior end of the hypothalamus. They’re their own thing.) State variables are things like hunger, temperature, immune system status, fertility, horniness, etc. I don’t have a great proof of that, just some indirect suggestive evidence. (Orexin, contiguity between lateral hypothalamus and PAG, various specific examples of people studying particular hypothalamus neurons.) Anyway, it’s hard to prove directly because changing a state variable can lead to taking immediate actions. And it’s really just a rule of thumb; I’m sure there’s exceptions, and it’s not really a bright-line distinction anyway. The literature on the lateral hypothalamus is pretty bad. The main problem IIUC is that LH is “reticular”, i.e. when you look at it under the microscope you just see a giant mess of undifferentiated cells. That appearance is probably deceptive—appropriate stains can reveal nice little nuclei hiding inside the otherwise-undifferentiated mess. But I think only one or a few such hidden nuclei are known (the example I’m familiar with is “parvafox”).

4sarahconstantin2y

plausible...but surely walking isn't "consummatory"? And turning on the DBS doesn't cause "automatic/involuntary" walking movements.

Steven Byrnes2y40

Yeah, the word “consummatory” isn’t great in general (see here), maybe I shouldn’t have used it. But I do think walking is an “innate behavior”, just as sneezing and laughing and flinching and swallowing are. E.g. decorticate rats can walk. As for human babies, they’re decorticate-ish in effect for the first months but still have a “walking / stepping reflex” from day 1 I think.

There can be an innate behavior, but also voluntary cortex control over when and whether it starts—those aren’t contradictory, IMO. This is always true to some extent—e.g. I can vol... (read more)

sarahconstantin's Shortform

8

sarahconstantin's Shortform

8