dash starting a list item
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. 1 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 responds to a dash at the start of a new line, the kind used to begin a bullet point in a list. Its negative side is weak and does not show a clear pattern. Turning it up gave wandering text that did not show list behaviour.
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.
-
dark brown (2).⏎- Length: 11 cm (2)⏎- 6 g (2)
-
bones and the tailbone (coccyx).⏎- Because this part of the spine bears the weight and stress of the trunk and upper
-
family members, or student partners.⏎- The students' stories can be compiled into a slide show or movie.<|endoftext|>Hyper
-
dulse is rich in iron and vitamin A.⏎- Kelp and arame are great sources of iodine and vitamin C.⏎Sea
-
5 metric tons.⏎- These short journeys alone could lead to melting around 1,326,
-
008).⏎- Check the tires. Have your wheels aligned and keep your tires properly inflated. Low
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: " the best way to make a good product, and the best way to make a good product." and I have been thinking about recently: " the best way to make a good product, and the best way to make a good product."
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
"leads" -- contacts or wires that connect to the circuit board.⏎Note the polarity. Much like a battery, a diode has a polarity. The negative side, or cathode, is usually represented by a bar, band or line along
-
are two such fuels. So, what are they, and how do they compare to oil?⏎• Chemical makeup: L
-
you want to read music, but want to know where to start? The first step should be to become familiar with the music notes.⏎In the Western system of music
-
bps down to $2/Mbps, the volume of traffic during the same period might grow from 6Gbps to 10Gbps. The net result of these simultaneous forces is a 25% increase in monthly transit fees,
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
- L0:178 suppresses
- L0:1224 suppresses
- L0:1301 suppresses
- L0:2015 weakly suppresses
- L0:25 weakly excites
- L0:2349 weakly excites
- L0:463 weakly excites
- L0:806 weakly suppresses
- L0:2029 weakly excites
- L0:2888 weakly suppresses
- L0:1959 weakly suppresses
- L0:189 weakly suppresses
- L0:1538 weakly suppresses
- L0:38 weakly suppresses
- L0:1327 weakly suppresses
- L0:2818 weakly excites
- L0:520 weakly excites
- L0:378 weakly excites
- L0:869 weakly suppresses
- L0:965 weakly suppresses
- L0:594 weakly suppresses
- L0:351 weakly suppresses
- L0:63 weakly excites
- L0:494 weakly suppresses
Feeds later layers
- L17:1766 weakly excites
- L3:177 weakly excites
- L17:2558 weakly excites
- L4:121 weakly excites
- L22:561 weakly excites
- L18:2114 weakly excites
- L14:954 weakly suppresses
- L20:2342 weakly excites
- L12:1738 weakly excites
- L5:719 weakly excites
- L11:3015 weakly excites
- L8:3048 weakly excites
- L8:830 weakly suppresses
- L4:380 weakly excites
- L27:1180 weakly excites
- L17:101 weakly suppresses
- L14:1347 weakly excites
- L5:2557 weakly suppresses
- L2:1995 weakly suppresses
- L21:1968 weakly excites
- L2:796 weakly suppresses
- L6:2878 weakly excites
- L8:168 weakly excites
- L22:3006 weakly excites
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
- enic
- extr
- ei
- //{⏎
- ensation
- 続ける
- anc
- nant
- xDF
- hread
Suppresses
- vere
- nder
- lobs
- seriously
- ��
- ogue
- ược
- służ
- 멘
- ழ