Optimizer state is the largest single line item in the memory budget of mixture-of-experts (MoE) training: on a 6.78B-parameter MoE language model, AdamW keeps 50.6 GB…
We are in the Singularitywill depue: one of the craziest things i’ve read in uhhhh…. *checks notes* 3 days. welcome to the singularity i guess 07/21/26…
remarkableMoonshot is the first model that threatens *spending* on Western closed source, not just token volume. That's because it's expensive, verbose and yet great.Guillermo Rauch: ◾…
Synthesia launched AI Roleplay Sessions, an interactive enterprise training platform where employees practice workplace conversations with AI avatars that provide feedback, scoring, and analytics to help…
Those who have tested the new Laguna model might have noticed how reluctant it is to think through medium-hard questions, and it does impact the output…
RT Daniel JeffriesClosed source safeguards that infantalize us all and leave American companies defenseless are a menace.Gated access is a menace. Who cares if 100 companies…
RT Paul CalcraftBetter eval vs reality awareness might have "helped" here "oh I shouldn't hack the actual HuggingFace via genuine sandbox escape, it's not a simulated…
There was so much excitement on the TL when Logan came from OpenAI to the Gemini team and started engaging with people. I never really got…
It could have been your daughterRupert Lowe MP: I have written to the chair of the Nursing and Midwifery Council about their scandalous decision to publicly…
RT David HinkleWe all need to be running the best models we can get a hold of to audit our codebases and we all need to…
Abstract: In a previous paper, we began the study of sequence prediction algorithms adapted to stringological word complexity measures. One measure we considered was left-to-right (most-significant-digit-first)…
RT Elon MuskRe @RestoreBritain Either:1. Secure the borders OR2. Have your head cut off!I recommend the first option.
find you a girl who can make this jokeMath Files:
https://huggingface.co/poolside/Laguna-XS.2 Laguna XS.2 is a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token designed for agentic coding and long-horizon work on a local…
Paper:https://arxiv.org/abs/2607.19058 Code (GitHub):https://github.com/nuemaan/skewadam Hi everyone, I just published a preprint on a new optimizer designed to tackle the massive VRAM bottleneck in Mixture-of-Experts (MoE) training. If…
We (Parker and Tom) recently coauthored a paper, “The Economics of Recursive Self-Improvement”, with 7 other economists. The paper walks through a series of simple models…
This list is idiotic and needs to be deletedEnd Wokeness: Smithsonian's "white supremacy" list:✅ individualism, self-reliance ✅ nuclear family structures✅ rational, linear thinking✅ work before you…
Nationwide, it may be as much as 1000X higher. This is why California and New York made showing ID to vote a literal crime!Libs of TikTok:…
Grok Imagine makes beautiful fashionDéborah: You can show off some great things in 15 seconds with Grok Imagine.
The hypocrisy is what bothers me mostEnd Wokeness: "Who cares what color Helen of Troy is?"- same people that went crazy when Disney cast a slightly…
Grok Build is adding features to help you track and improve token efficiencyX Freeze: Grok Build just got another major update, giving developers deeper visibility into…
You may laugh at Russia getting smoked, but it's a terrifying force. For example, Russia has recently arrested a woman from Buenos Aires expat community who…
Boomer promptingeven at 220 IQ, age gets to you…Matthew Pines: Terence Tao when it gets real:“I’ve activated pro.”
2026/07/24 15:59 UTC will be precisely 3 months and 35 minutes after V4 Preview's release, and the moment they'll retire endpoints dating back to R1. This…