number word before a list of reasons
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. 17 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 words like 'two', 'several' and 'a few' when the text is about to list a set number of reasons, ingredients or ways. Turning it up did not produce a clear, related change in the output.
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.
-
We have a centuries-old accumulation of armaments that shows no sign of evaporating.⏎But there are two things that are needed for a gun to work: the gun and the ammunition.
-
city of⏎Gundisapur in 'Ylam at the eastern bank of today's Shatt-il-Arab.⏎These remarks indicate two ingredients in the formation of early Mesopotamian⏎Christianity, that have
-
and that of human civilization go hand in hand.⏎In fact, historians believe that yogurt played a pivotal role in the development of human societies. This is due to two main reasons. First, by allowing milk to be stored for longer, more people
-
⏎- Door. …⏎- Lower Door Skin.⏎Why are car bumpers made of plastic?⏎Plastic is the preferred material for modern car bumpers for several reasons. … First and foremost, plastic is much lighter and more aerodynamic than
-
blog post we will use a Python script to work out how many lego bricks would be needed to build planet Earth!⏎To solve our problem we will use the following data:⏎We will work out the number of lego bricks needed in several step
-
, given up to an animal rescue group, or put to sleep.⏎Compared to other states’ cruelty laws, Texas animal cruelty laws are unique in a couple of ways. First, Texas criminal cruelty laws only apply to domesticated animals, such
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 can't get over the fact that the (time) (time) (time) (time) (time) (time) (time) (time) (time) (time) (time) (time) (time)
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
has a lot of items that can’t be moved, then a portable play yard is a good idea–a place where he can handle anything he finds. But again he should have new things to discover every day.⏎The word “gentle”
-
in Joules⏎c) The amount of heat added to the gas in Joules⏎d) The change in energy of the gas in Joules⏎I've
-
honors courses?⏎You must have come across this term while searching through college websites. Basically honor courses are a “distinctive
-
, and pigment granules. Understanding these components will not only deepen your knowledge of hair but also help you identify them and comprehend their significance. So, let's dive in!⏎The cuticle is the outermost layer of the hair shaft, serving
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
- L16:655 excites 'see' cross-references and web boilerplate
- L16:1961 suppresses
- L15:427 excites
- L15:2777 suppresses
- L14:1738 excites
- L16:2539 suppresses
- L16:888 excites
- L15:717 suppresses
- L15:2158 suppresses
- L16:1127 suppresses opening bracket after an imperial measurement
- L15:2258 excites
- L13:3004 suppresses
- L15:1119 suppresses
- L14:667 excites
- L15:695 suppresses
- L16:737 excites
- L15:2145 weakly suppresses
- L15:1441 weakly excites
- L15:2871 weakly excites
- L5:1039 weakly suppresses
- L16:2665 weakly suppresses
- L16:1680 weakly suppresses
- L15:2610 weakly suppresses
- L11:513 weakly excites words inside idioms and fixed phrases
Feeds later layers
- L19:2149 excites
- L18:548 excites
- L18:1483 excites
- L19:2239 excites
- L18:2261 suppresses
- L19:2727 excites
- L21:2585 excites
- L18:2843 suppresses
- L18:2572 suppresses
- L18:2359 excites
- L18:468 weakly excites
- L20:2268 weakly excites
- L20:3049 weakly excites
- L21:2353 weakly excites
- L19:2843 weakly excites
- L18:132 weakly excites
- L26:2817 weakly excites
- L20:2127 weakly suppresses
- L18:3046 weakly suppresses
- L18:959 weakly suppresses
- L21:2273 weakly excites
- L23:1409 weakly excites
- L21:2256 weakly excites
- L18:731 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
- 几种
- 事情
- oth
- var
- 有一些
- 几个
- ric
- 有所不同
- 有所
- 两种
Suppresses
- TemplateName
- qrstuvwxyz
- ”⏎⏎
- agger
- ”⏎⏎
- _Default
- Template
- CPP
- .assertNot
- ({⏎