"whatever you now find weird, ugly, uncomfortable and nasty about a new medium will surely become its signature. CD distortion, the jitteriness of digital video, the crap sound of 8-bit - all of these will be cherished and emulated as soon as they can be avoided."
- brian eno
New neurons research: introducing Neuron Communities.
We discovered communities of neurons that work together inside LLMs. They’re surprisingly small, but powerful – just a few neurons work together to perform common behaviors, like arithmetic, ALL CAPS, or recalling a date.
Work led by research fellow @_igorshilov:
It would be fascinating to take open base models from 2023-2024, apply modern midtraining and RL to them and rerank their capabilities.
I bet there were some real sleeper bangers in there that were just a good post-training and reasoning capabilities away doing cool stuff.
Can an LLM understand what changing another model's weights will do without ever running that model?
A new experiment called Weight-Space Reading explores this fascinating possibility using open-weight Qwen and Muse models ranging from 9B to 30B parameters.
Some interesting findings:
73% balanced accuracy: Qwen 27B predicted subtle weight-edit effects using direct weight access, compared with 57.8% using text descriptions.
Self-inspection remains uncertain: No conclusive advantage was established for models reading their own weights.
False predictions: On Muse-30B, 34 standard edits and 6 no-op controls produced no detectable behavioral changes, yet agents frequently predicted changes.
The researchers have also released an evaluation harness, experimental code, and an open-source toolkit for weight-space programming.
The bigger question: Could LLMs eventually inspect, understand, and predict the behavior of neural networks directly from their parameters?
This could open an entirely new direction in model interpretability and automated weight editing.
GitHub: github.com/suayptalha/wsr
Paper: alphaxiv.org/abs/2611.weight…
casually remote controlling an orchestrator agent that delegates across a team operating on remote GPUs over the weekend and having preferences noted and routine workflows distilled into invokable skills is such a futuristic kind of convenience of u pause to appreciate it
Reprogramming dormant neurons : Qwen TTS
Ever wondered why Qwen3-TTS sometimes breathes or laughs out of nowhere? Agent found neurons behind it.
The details:
Agent found 417 neurons to be specific (about 0.48% of the talker's MLP neurons) and 61 of them are dormant. Amplified together, they give the following effects effect:
1. Breathes as words are spoken
2. Hushed ASMR
3. Natural laughs as language is being spoken
4. Enhanced whispering
5. 2nd speaker appearing.
Comparisons and neuron map repo in the comments below.
Inspired by recent work on Lazarus by @DimitrisPapail , which reprogrammed dormant neurons.
i'm really not a big fan of "invisible nontransparent harness bullshit" and Anthropic appear to be really big fans of "invisible nontransparent harness bullshit"
You’re right. The VHF (Very High Frequency) band from 118.000 to 136.975 MHz used for civilian aircraft communication shouldn’t be reprogrammed to play iconic hits of 1970’s classic rock. And that’s on me.
There has been a temporary telecommunications outage - out of an abundance of caution all flights have been halted. Every flight has been accounted for and international flights will be handled on a case by case basis.
masked decoding on gemma 3 12b to control text rivers
if a token isn't the length i want it to be, i mask it out
added a lora for width tags so the model is aware of text length + coherence RL afterwards