Meme-antine
Some links, as usual
Big LLMs are converging on their world model?
A peculiar investigation into how large models are slowly (ahem AI progress ahem) moving towards a shared, singular statistical world representation. It’s peculiar in a “is that the true reality model, or just the most common in human knowledge?” sense as the models were trained using different datasets, yet arrive at the same (vector?) conclusions.
The closer a model is to Meta’s DINOv2 vision model, the better it performs 👀 For some reason, I get the proximity-based truth discovery (~by great great Omer Ben-Porat I’ve had the privilege to chat with a couple of times) vibes from these concepts.
Cancer and obesity
Original preprint is here.
5-point increase in BMI increased the risks of developing cancer by 12% for women and 24% for men;
Obesity and BMI influence at least 31/122 types of cancer, and account as a factor in ~40% of overall cases;
The study involved 4.1M individuals and 332K cancer cases, over 100M person-years of follow-up (~25 years per per, son…)
Haha, squamous cell carcinoma risks are lowered with weight! Eureka!
Just some images…
Kidney != invincibility :(
Richard Slayman, having an end-stage renal disease, agreed to have a xenotransplantation (xeno- = foreign, different) of a modified pig kidney but, unfortunately, passed away two months later :(
Still, it’s a big step towards this practice (xenotransplantation, not death…) becoming widespread. Hats off to the mister 💗
Surprise, surprise. Wait, who are you?
APOE4 gene, thought to increase the risk of Alzheimer’s, seems to be the cause in some of the cases - provided the someone has two copies of it.
Here are some second-order pathways of APOE4’s ⬆️ ⬇️:
Knowledge graphs out of thin text
Rahul Nayak had made a nifty library for creating knowledge graphs given an ontology (hierarchy of data and relations). Before that, he had made a post on doing the same without an ontology, letting LLM decide the best one.
As usual with all things neurosymbolic, IBM has a body of research covering the topic of KGs.
Memes and ants
In rats, 21 days of imipramine causes D2 receptor supersensitivity ➡️ mania-like syndrome, followed by an opposite episode when imipramine is discontinued;
This phenomenon is well-documented, including humans - basically an unlucky genetic composition and a course of ADs can switch one from unipolar to bipolar spectrum;
Stimulation of NMDA receptors is required for the sensitization to occur, as MK-801 blocks this phenomenon;
Lithium, carbamazepine and valproate do not prevent this from happening, but memantine does, preventing ‘mania/depression’ rollercoaster;
Memantine is unique in the regard it prevents excitotoxicity in extracellular receptors but doesn’t interfere with normal synaptic NMDA activity, like ketamine and MK-801.
- Bipolar disorder (BD) has a lifetime prevalence of approximately 5%. 83% of BD cases are classified as “seriously severe” and 17.1% as “moderately severe”;
Giving MK-801 and ketamine to your mice is a great way to drive them miceshit crazy!
Aforementioned extracellular NMDA effect gives memantine neuroptophic properties. It also shares some awesome properties with lithium: ⬆️BDNF, ⬇️PKC and ⬇️GSK-3b;
Memantine had also shown promise for ADHD, anxiety, sleep and BPD management;
Looks like the extracellular blockade is mostly present with low-dose memantine (although not shown in humans except a pilot study, likely to extrapolate well)
Yeaah, low-dose 🌞
Welcome to Teleogenic/Boi Diaries❣️
Other places I cross-post to:
Reuse
Citation
@online{kogan2024,
author = {Kogan, Zakhar},
title = {Meme-Antine},
date = {2024-05-19},
url = {https://teleogenic.com/posts/240519-memantine/},
langid = {en}
}