the word 'wide' in 'a wide range'
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. 2 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 fires on the word 'wide' in phrases like 'a wide range of' or 'a wide variety of'. Its negative side is weak and does not show a clear pattern. Turning it up made the model write about many different topics, but not in a way that is clearly tied to the word.
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.
-
offers a wide variety of Online TEFL Courses and a great number of opportunities for English Teachers and
-
able long-term to examine the health effects. Researchers have not been able identify all possible chemicals in ecigarettes.⏎Several studies have shown that vaping can cause a wide range of short-term health problems. You may experience inflammation in your lungs, cough
-
diseases is only anticipated to rise as the population ages. The same is valid internationally and increasingly in emerging nations.⏎These trends have sparked a wide range of global, national, state, and local initiatives, as well as in
-
, such as factories, may also be at risk of exposure to rat killers, slug pellets, ant killers and weed killers.⏎Pesticide poisoning can present a wide variety of symptoms, depending on both the type of pesticide that a person is exposed
-
7 to Year 10, but preferably in Years 7 or 8.⏎In Years 11 and 12, NSW schools offer a wide variety of languages, catering for beginning students to background speakers.⏎Our Faculty Members⏎
-
Great Lakes water quality, rapid reduction in ice cover and the resurgence of some pollutants like excess nutrients are among the indicators currently raising concerns. Based on contributions from a wide range of U.S. and Canadian experts, the report reveals a
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: I feel like I'm trying to be a part of the world. I'm trying to be a part of... what I'm. I feel like I'm trying to be... I feel like I'm trying to be... I feel like
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, adaptable, and easily taught.⏎“The social relationships [between] these things matter a lot. And these things have got to be created, developed
-
lateral sclerosis (ALS) and Huntington’s disease, but don’t always activate the right proteins at the right time, a recent study shows.⏎“[The cells] do not always realize there is a problem, or know how to solve it,
-
of meteor surveillance cameras to determine the fireball's trajectory, denoted by the black arrow in the map below:⏎"The asteroid entered at a [relatively slow] speed
-
pes simplex virus] and other pathogens," the researches wrote in their report. "Circum
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
- L1:1833 weakly excites
- L1:809 weakly suppresses
- L1:1478 weakly excites
- L0:933 weakly excites
- L1:222 weakly excites
- L0:175 weakly suppresses
- L0:2813 weakly suppresses
- L0:385 weakly excites
- L1:1971 weakly suppresses
- L1:175 weakly excites
- L0:834 weakly suppresses
- L0:201 weakly suppresses
- L0:2067 weakly suppresses
- L1:1016 weakly excites
- L0:2349 weakly excites
- L0:1661 weakly suppresses
- L0:2216 weakly excites
- L1:130 weakly excites
- L0:877 weakly excites
- L0:2543 weakly excites
- L0:376 weakly excites
- L0:92 weakly suppresses
- L0:1847 weakly suppresses
- L0:2669 weakly excites
Feeds later layers
- L15:1667 weakly excites
- L21:2058 weakly suppresses
- L8:1112 weakly suppresses
- L25:1161 weakly excites
- L15:936 weakly excites
- L8:2346 weakly suppresses
- L6:1043 weakly suppresses
- L27:307 weakly suppresses
- L4:2132 weakly suppresses
- L5:1086 weakly suppresses
- L11:1683 weakly excites
- L15:257 weakly suppresses
- L15:2373 weakly excites
- L20:1352 weakly excites
- L17:1268 weakly suppresses
- L22:2756 weakly suppresses
- L25:1760 weakly suppresses
- L9:1182 weakly suppresses
- L6:939 weakly excites
- L13:787 weakly suppresses
- L13:1516 weakly suppresses
- L22:85 weakly suppresses
- L19:2907 weakly excites comma right after 'and' in lists
- L9:2283 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
- firstname
- accord
- represented
- ouncill
- need
- .Display
- Needs
- destruct
- 化进程
- acceptable
Suppresses
- ler
- kicked
- RESS
- onyms
- lers
- inois
- ellen
- emen
- gens
- ression