Лига Хруща // League of Hrusch

😭

>Открываю Google AudioSet Dataset >Как поднять ФПС и снизить пинг в танках

What most people don't understand about the significance of Grok 4.5/Muse 1.1 release compared to Fable - Anthropic are not earning money with those models. Their business model is not sustainable. They are a non-profit burning off VC money. The real usable models are a tiny bit (3-5%) worse, while being 10 times cheaper - and guess which companies exist for decades and have both the experience and, most importantly, THE DISTRIBUTION to reach their audiences? Also, 99% of model use-cases are not coding, CyberSec and SWE are not the ones bankrolling their development

But...

Вышел Muse 1.1 Muse 1.1 is $1.25/4.25 per input/output tokens https://x.com/alexandr_wang/status/2075218936266998230 https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/

META IS SO FUCKING BACK

Got Qwen 3.6 A3B running locally on a Mac - surprised by the output speed. A greatly usable local model

Sneak peek into what I was working on lately - it is an annotation tool for medical conversations, that allows to extract clinically relevant entities (drug dosages, symptoms, treatment plans, etc.) based on a conversation. Each entity is a node in a graph, they can be related to multiple other entities, and they point to specific spans in the text, so if you click on the annotated entity, it will show specifically where it was mentioned. Of course, there is also an option to generate the annotations automatically using an LLM :D

https://luma.com/cerebras-piwl

UN document released in year 2000 https://www.un.org/development/desa/pd/sites/www.un.org.development.desa.pd/files/unpd-egm_200010_un_2001_replacementmigration.pdf

lulw

holy quantized slop for 200 USD Claude

когда уехал от жары:

Solo author, one 3090ti, and a dream https://arxiv.org/abs/2512.08309

Insight: When you have a strong urge to buy something, buy stocks You will get the same sense of joy from wasting money, but it will not hurt you long term 😜