nic 🔹
banner
nmk.wtf
nic 🔹
@nmk.wtf
mathematician, aspiring quant, reader of things
https://nmk.wtf
I give 10% of my income to GiveWell and so should you
The 50 year mortgage is an intrinsically toxic product
byrnehobart.medium.com/the-30-year-...
The 30-Year Mortgage is an Intrinsically Toxic Product
The crash didn’t pop the bubble
byrnehobart.medium.com
November 11, 2025 at 6:48 PM
The blog post for people who are interested:
pleias.fr/blog/blogsyn...
News & Thoughts
We build LLMS with advanced reasoning capabilities for information-intensive and highly regulated industries
pleias.fr
November 11, 2025 at 2:39 PM
The original tweet is Elon. MattY is quote-tweeting Elon. Quote posts look the same here on bsky.
November 10, 2025 at 11:45 AM
November 9, 2025 at 3:25 PM
I just also want to say that this line of reasoning could be quite fruitful! Similar thinking gave us shampoo (which is apparently widely used (or at leas variations of it)): arxiv.org/pdf/1802.09568
arxiv.org
November 6, 2025 at 3:51 PM
The reason to try it is to maintain some invariance, Like normally we want the pre-activations to be ~normal. So maybe you want the pre-activations to be ~Gaussian Ensemble (¯\_(ツ)_/¯)
November 6, 2025 at 12:05 PM
This would be less expressive compared to just writing all of those matrix entries in a vector. You would have to think carefully about what invariance you are maintaining!
November 6, 2025 at 11:51 AM
Clarification: In this scenario do you envision your weights to just be 2d matrices as well?
November 6, 2025 at 11:49 AM
This seems to generally be a good idea for Pigouvian Taxes, give the money back to the people!
November 3, 2025 at 3:35 PM
Okay interesting. Poking around in the hugging face I only found the embedding matrix. Might be a huggingface thing?
November 3, 2025 at 1:56 PM
Dumb question: When I'm on the Qwen 3 HuggingFace I don't see the unembedding matrix? Where is it? I have checked the various `
model-0000x-of-00003.safetensors` files huggingface.co/Qwen/Qwen3-4...
Qwen/Qwen3-4B-Instruct-2507 at main
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
November 3, 2025 at 1:02 PM
Politics comes second. First we will have the experiment. Regardless of "intelligence", these models will affect huge changes in unforeseen ways (as they already do!). My guess is that AIs *will* manage workers before anyone really thinks about what that will mean or entail (probably for the worse)
November 2, 2025 at 3:46 PM
Information hygiene is extremely important in the age of letting tons of random garbage into our brain every second
November 2, 2025 at 10:00 AM
lerw_and_loops.py
GitHub Gist: instantly share code, notes, and snippets.
gist.github.com
November 1, 2025 at 11:00 AM