comma after sentence-opening adverb
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. 0 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 comma that comes right after a linking word that starts a sentence, such as Consequently, Actually, Indeed or Later. When it was turned up, the model's text did not visibly change in a way that matches this pattern.
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.
-
also just transition policies.⏎Consequently, in order for climate action to be successful, indigenous peoples must be seen as powerful
-
a shield.⏎Due to the widespread use of shields by all Carolingian troops, it became the most vital defensive armour for Carolingian armies.⏎Consequently, Carolingian monarchs such as Charlemagne specifically ordered different monasteries
-
differentials.⏎Actually, there has been no shortage of explanation. Going through some of the literatures,
-
of “Seeds of Deception”, in one of the first studies in the early 1990’s, rats were fed GM tomatoes. Actually, they refused to eat them, so they had to be force fed. And,
-
to industrial robots, they are programmed to operate at lower speeds and with less ”torque strength”, making them safer to use in parallel to human workers.⏎Indeed, industrial robots are often used for larger sized tasks that require bigger installations to fulfill the
-
data the supervisor of the construction was Bela Colov.⏎Later, the tower had a defensive character as well, especially in the time of the attack
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 thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
subsequently destroyed by Danish invaders in the 11th century.⏎The wave of religious
-
Thus, a system may consist of neural networks named Alice and Bob, and we aim
-
Thus, the increasing demand for energy and exponential surge in consumption stimulates problems such as an
-
thereby triggering a T cell immune response locally that then attacks cancer throughout the body. We
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
None: this is the first layer.
Feeds later layers
- L2:1283 weakly suppresses
- L3:1984 weakly suppresses
- L3:845 weakly excites
- L7:1776 weakly suppresses
- L6:427 weakly suppresses
- L21:1096 weakly suppresses
- L10:2550 weakly suppresses
- L6:1760 weakly suppresses data quality and data formatting
- L15:716 weakly excites
- L8:1197 weakly suppresses
- L18:1678 weakly excites
- L4:1078 weakly suppresses
- L5:2509 weakly excites
- L22:415 weakly excites
- L12:766 weakly suppresses
- L18:2548 weakly excites
- L24:1650 weakly suppresses
- L23:478 weakly suppresses
- L1:2069 weakly suppresses
- L7:1809 weakly suppresses
- L8:1399 weakly suppresses
- L7:2854 weakly suppresses
- L19:2514 weakly suppresses
- L25:1532 weakly suppresses 'New' in place names
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
- ongs
- 了许多
- StateManager
- ителя
- ителей
- essor
- ister
- 咛
- ynet
- .deg
Suppresses
- ocket
- 发现问题
- tesy
- 目的是
- **)&
- romise
- hecy
- 以此
- 셋
- 反应