"University" in affiliation names
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 21 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron reacts to the word "University" when it names a person's workplace, like "professor at the University of Florida". It is mostly quiet on other words, and the steered text didn't show this theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).
Fires most on
The text windows where this neuron's activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
Jordan M Prutkin, MD, MHS, FHRS⏎- Associate Professor of Medicine, Division of Cardiology, Electrophysiology Section⏎- University of Washington⏎- Leonard I Ganz, MD, FHRS, FACC
-
of Philosophy at the University of Texas at Arlington.⏎—Lawrence Zbikowski, Department of Music, University of Chicago⏎—Anthony Cascardi, University of California, Berkeley⏎—Jesse J. Prinz, John J.
-
signs of hoarding disorder.⏎The short answer is: Probably not. Hoarding disorder goes beyond stockpiling in an emergency. I am a psychiatrist at the University of Florida and the director of the Center for OCD, Anxiety and Related Disorders.
-
critically ill children when there are no pediatric specialists available in their own communities,” says James Marcin, director of the Children’s Hospital Pediatric Telemedicine Program at the University of California, Davis.⏎“People say a picture is worth a thousand words,”
-
familiar?⏎DAVID GROSS (University of California, Santa Barbara): Does it bother us? Absolutely.⏎STEVEN WE
-
’65, Peggy Phelps, Jane Hurley Wilson ’64 and Michael G. Wilson, and the Getty Foundation.⏎Daniela Gonzalez-Pruitt, University of Toronto ’17 and Josephine Ren, SCR ’19⏎Stories
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.
Here is something I have been thinking about lately: what if I have the same number to the right and the same number to the rightmost number, and the same number to the rightmost number. So, I have a normal $3$-normal number, which can be considered a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
effective border biosecurity and to mitigate and prevent the spread of biological threats throughout the region. However, a 2022 study by the Hanoi University of Public Health and the Global Initiative Against Transnational Organized Crime (GI-TO
-
Mining-industry; Underground-mining; Coal-mining; Mine-safety; Ventilation; Coal-dust; Dust-control; Control-technology⏎National Institute for Occupational Safety and Health, Pittsburgh Research Laboratory, Pittsburgh, PA 152
-
other experiences, Vladimir says he's determined to carry on spreading the message about HIV.⏎Jaroslav Jedlicka, AIDS Programme Manager at Prague's National Institute of Public Health, says awareness is relatively high in the Czech Republic. However, he
-
can learn near optimal policies based on trial and error interaction with the . learning objectives samples. carnegie institute of technology. chemical reaction engineering; feedback control systems. college of fine arts. architectural
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L10:3060 suppresses
- L2:2708 weakly suppresses
- L20:1215 weakly suppresses
- L11:1994 weakly excites
- L17:2075 weakly excites
- L4:2677 weakly suppresses
- L19:2246 weakly suppresses
- L19:212 weakly suppresses
- L7:2662 weakly suppresses
- L13:2104 weakly excites
- L9:903 weakly suppresses
- L12:1969 weakly excites
- L1:37 weakly suppresses
- L9:88 weakly excites
- L18:1110 weakly suppresses
- L12:1598 weakly suppresses
- L15:3043 weakly excites
- L20:1254 weakly suppresses
- L18:2292 weakly suppresses
- L4:2123 weakly excites
- L20:1887 weakly suppresses
- L6:2855 weakly suppresses
- L7:180 weakly excites
- L7:99 weakly excites
Feeds later layers
- L22:2000 weakly suppresses
- L22:1016 weakly suppresses
- L22:172 weakly suppresses
- L24:2816 weakly excites
- L25:2559 weakly excites
- L22:1603 weakly excites
- L24:1236 weakly suppresses
- L24:1757 weakly suppresses
- L27:864 weakly excites
- L22:2153 weakly excites
- L24:2937 weakly suppresses
- L24:2426 weakly excites
- L27:2815 weakly suppresses
- L24:1831 weakly suppresses
- L26:1507 weakly excites
- L24:697 weakly suppresses
- L22:1483 weakly excites
- L26:1020 weakly suppresses
- L24:2818 weakly suppresses
- L22:493 weakly suppresses
- L24:1172 weakly suppresses
- L22:659 weakly suppresses
- L24:454 weakly suppresses
- L27:267 weakly suppresses
Activation histogramA chart of how often a neuron takes each activation value.
How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- →Array
- Array
- Array
- (Array
- .ibm
- Arrays
- ropy
- arrays
- ARRAY
- synagogue
Suppresses
- 심
- eming
- #$
- 技术水平
- Processes
- sammen
- _duplicates
- 的程度
- _processing
- /archive