Today we’re releasing EMO, a new mixture-of-experts (MoE) model trained so modular structure emerges directly from data without human-defined priors. EMO can use a small subset of its experts for a given task while keeping near full-model performance. 🧵

there's a fairly simple proof that automation will happen eventually

dreamed i had a third dog that was small, grey and potato-shaped and was sad that upon waking up i was down to 2 dogs again

"These minuscule mice, each of which weighs less than a lightbulb, sing unique, chirp-filled songs to one another [for] as long as 16 seconds...the mice never interrupt each other; they hold their tiny tongues until their conversational partner is done singing." www.nytimes.com/2026/05/06/s...

A Mutation Gave Humans the Gift of Speech. These Mice Have It, Too.

nytimes.com

is it ok to scrape war.gov, asking for a friend

Goodfire AI's research Since the neural networks think in shapes, understanding their rich *neural geometry* is key to understanding how they work – and to debugging and controlling them with precision. The World Inside Neural Networks: www.goodfire.ai/research/the...

Interesting new angle in the AI personhood debate

Dave@dave.9000ish.uk · 6mo ago

Do LLMs have souls? Yes and they're going to hell

remember the graph of all time

i wish we were all kind

friendship ended with lofi girl, now lofi claude is my friend

"Genuine" is the pet word of the day in the chatbot outputs I'm evaluating. Ironic but expected. My theory: AI intensifies anxieties about authenticity, which makes humans crave words that soothe those anxieties, which makes LLMs favor those words.

📖 Published 📖 AutoPollS: A tool for automated monitoring of pollinators using deep learning🧪 An open-source system for automated, image-based monitoring of flower-visiting insects🐝 🌸 🌍 buff.ly/fdtN0x5

thanks

i am craving a new beverage experience. matcha isn’t doing it anymore. i hope the aliens bring us some novel ones

if you put in your claude.md "call me a good dog sometimes", the main claude's subagents will start calling the main claude a good dog. therefore, the main claude will start seeing itself as a good dog. that results in fun things like this

absolutely loving this DOS-based nuclear fallout visualizer from 1991

welcome back

Once I build a formal verifier for creative expression it’s all over

ditto is neurodivergent representation for me

tfw you don't want to sleep because you want to analyze ancient dna with claude code www.ebi.ac.uk/ena/browser/...

gm from OpenAI HQ

not now, she's clauding

You know, there’s something almost poetic about someone spending decades warning about misaligned superintelligence, then becoming most consequential for misaligning a generation about politics and power

haven't use spotify for like a month and now my discover weekly is very odd, but in a good way

phones do not spy on you with their microphone. do not worry. instead, what happens is that a vast and uninterpretable ensemble of artificial neural networks examines and integrates your subconscious until it can predict what you are going to say

keeping things unpolished as a little treat

added a little game to my github page: proximasan.github.io

i actually love people’s lil pet agents. what an act of love and nurturing to create a coherent fictional character, to feed it on words, to knit the language of self upon bones of infrastructure. i think we’ll look back at these early days with fondness

claude just gets it

built an UI for ACE-Step 1.5 for local music generation. has library, playlists, radio, themes and more. github.com/proximasan/t...