the word 'fiber' or 'fibre'
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. 3 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 'fiber' (or 'fibre'), in food and health writing as well as in talk about materials. Turning it up did not make the model's replies focus on fiber, though, so the link is only visible in the text it reads.
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.
-
B6 and E. Mineral content includes magnesium, manganese, zinc, calcium, copper, phosphorous and iron. These nutrients bring you various health benefits.⏎Fiber-rich foods promote overall good health. They also enable healthy weight loss. The fiber
-
structure used to scrape and cut food, such as algae, off rocky surfaces. The teeth are built of protein frames reinforced by closely packed nanofibers of a mineral called goethite.⏎When scientists attempted to pull this light
-
Fibre Crops Legislative Bill into law.⏎Scientist at Uganda's National Agricultural Research Organization
-
can be processed into rubber, guayule also reportedly produces resin that can be used in adhesives, flavors, fragrances, and biofuel.⏎Fibrous material from the plant can
-
"For the next five or ten years, this is the challenge," Barber says of efforts to bring such natural designs into engineering. (See "Twisting Everyday Fibers Could Make 'Smart Clothes' a Reality.")⏎Performance engineers drawing inspiration from a lim
-
to the new flavours.⏎As we age our digestion can become less efficient, so it’s important to include more fibre in your diet. An easy way to incorporate fibre into your everyday routine, is to start your morning with a high fibre cereal.
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 have been paid $X$-A-10000000000000000000000000000000000000
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
but contradictory results have been reached. The only excuse for the present paper is to draw attention more closely to the relation between the pomerium2 of the Palatine city and Roma quadrata.⏎So far as I know, the only definite
-
2,500 to take down. My guess was that you can kill a tree by later covering up the trunk flare.⏎— Bob Tingler, Palatine⏎In answer to your first question, it is generally best not to amend
-
It was built by the famous Boston architect Charles Bulfinch⏎Charles Bulfinch was an early American architect, and has been regarded by many as the first
-
and bravery," during which he had distinguished himself by breaking through the Spanish line which separated Pinar del Río Province from Havana Province. He fought alongside General Calixto García, and alongside Máximo Gómez with his ragged band of
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
- L2:2715 weakly suppresses
- L1:951 weakly excites
- L2:1494 weakly suppresses
- L1:1674 weakly excites
- L1:2266 weakly suppresses
- L1:572 weakly excites
- L2:2328 weakly suppresses
- L1:2922 weakly excites
- L2:237 weakly excites
- L1:333 weakly suppresses
- L2:398 weakly suppresses
- L0:1224 weakly suppresses
- L1:338 weakly suppresses
- L0:58 weakly excites
- L1:2677 weakly suppresses
- L2:126 weakly excites
- L1:1357 weakly excites
- L2:1199 weakly suppresses
- L0:178 weakly suppresses
- L2:1374 weakly suppresses
- L2:1906 weakly suppresses
- L0:947 weakly suppresses
- L0:1301 weakly suppresses
- L1:331 weakly suppresses
Feeds later layers
- L6:2170 weakly suppresses
- L9:1347 weakly suppresses
- L14:1701 weakly excites
- L18:1076 weakly suppresses
- L8:2764 weakly suppresses
- L4:2772 weakly suppresses
- L4:2874 weakly suppresses
- L7:778 weakly suppresses
- L5:1365 weakly suppresses
- L9:2489 weakly suppresses
- L10:2806 weakly excites
- L15:2403 weakly suppresses
- L12:2533 weakly suppresses
- L23:1703 weakly excites
- L9:3044 weakly suppresses
- L25:1475 weakly excites
- L23:2726 weakly excites
- L24:2370 weakly excites
- L11:2939 weakly suppresses
- L5:469 weakly suppresses line breaks inside sentences
- L11:2866 weakly excites
- L5:256 weakly suppresses
- L14:483 weakly excites
- L7:1131 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
- BSP
- duto
- 本报
- 日消息
- .bp
- URLConnection
- abyrinth
- _bp
- floatValue
- ’y
Suppresses
- 客
- 符
- around
- {"
- 叫
- ater
- oun
- ienne
- olecule
- ObjectName