dash starting a new 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. 5 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 a dash that starts a new point in a list or a fresh item right after a line break. Its negative side is weaker and mostly fires on opening quotation marks. Turning it up did not produce an obvious change in the replies.
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.
-
-world problems involving geometric sequences.⏎Terms introduced in this lesson:⏎Teaching Strategies and Tips⏎Use this lesson to show how exponential functions and geometric sequences are related.⏎- In a geometric sequence, terms are found by multiplying the same constant to the previous
-
and silent killer. In the time it takes to do these simple tasks, a child can become submerged, loose consciousness, sustain permanent brain damage, or die.⏎- Get a towel (10 seconds)⏎- Answer the phone (2 minutes)⏎
-
= acceleration due to gravity⏎Since gravity is just an acceleration like any other, it can be used in any of the formulas that we have used so far.⏎- Just be careful about using the correct sign (positive or negative) depending on the
-
two.⏎Serious Causes of Eye Redness⏎Redness of eyes with pain, loss of vision, inflammation or thick discharge may necessitate an immediate treatment.⏎- Uveitis. This is the
-
atively.)⏎DiffuseThe adjective diffuse means spread out or not concentrated in one place. Diffuse can also be a verb meaning to circulate or to spread.⏎- She is the CEO of a large diffuse company. (Here, diffuse is an
-
likely will not notice anything odd when mites are at low populations; however, when they continue to grow to a high population you will start seeing the damage.⏎- Stippling of light dots on leaves⏎- Leaves turning yellow or reddish
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 know that the book is in the public domain. The user has no idea what's next. I'm not sure about the number of words. It's a book I'd like to be read by the people who are left in the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
gender-based violence, and challenging harmful stereotypes faced by people of all genders, we are creating a foundation for everyone to lead happier, healthier and more fulfilling lives.⏎“Like Canadians, our government believes that
-
Revolution, it easily identifies with the struggles and aspirations of working women everywhere. The figures represent women of different races and celebrate the contributions made by women throughout time.⏎“Women all over the world have one trait in common: they work, they work
-
.⏎“Particularly in low-income countries, exhausted health workers are battling to save lives in the midst of shortages of personal protective equipment, oxygen, and treatments.⏎“Vaccines have never been the way out of this crisis on their own,
-
, and to manipulate it. ... To arrange atoms into an ordered nanotechnology landscape of precisely spaced steps and terraces. ... To determine the size of an electrical current by tabulating, one by one,
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
- L3:44 weakly suppresses
- L2:308 weakly suppresses
- L1:107 weakly suppresses
- L4:140 weakly suppresses
- L2:163 weakly excites
- L2:2396 weakly suppresses
- L4:231 weakly excites
- L1:1648 weakly excites
- L0:953 weakly suppresses
- L2:45 weakly excites
- L2:266 weakly suppresses
- L3:141 weakly excites
- L1:302 weakly suppresses
- L4:1175 weakly suppresses
- L3:1404 weakly excites
- L3:2208 weakly excites
- L4:1749 weakly excites
- L3:2567 weakly suppresses
- L2:329 weakly excites
- L3:1093 weakly suppresses
- L2:111 weakly suppresses
- L4:841 weakly suppresses
- L2:2550 weakly suppresses
- L3:145 weakly excites
Feeds later layers
- L6:172 suppresses
- L8:57 suppresses
- L10:801 excites
- L6:164 excites
- L6:1704 suppresses
- L7:1347 excites dash starting a new list item
- L7:237 suppresses
- L11:548 suppresses
- L8:2937 excites
- L17:84 excites
- L6:445 suppresses
- L12:1483 weakly excites
- L15:986 weakly suppresses
- L16:1389 weakly suppresses
- L16:300 weakly suppresses commas inside lists of nouns
- L13:371 weakly excites
- L8:299 weakly excites
- L8:131 weakly excites
- L7:25 weakly suppresses
- L8:739 weakly suppresses
- L15:215 weakly suppresses
- L12:399 weakly suppresses
- L24:1993 weakly suppresses
- L15:2287 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
- zman
- 也不例外
- arness
- sterol
- ETHER
- 局副局长
- erties
- acias
- 局局长
- passed
Suppresses
- urance
- 룸
- olo
- rowable
- hou
- odable
- nop
- oub
- ствие
- 问题是