'the' after 'member of' or 'one of'
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. 7 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 word 'the' when it comes after phrases like 'one of' or 'members of', where a group is about to be named. Turning it up gave repetitive or odd text with no clear link to the group idea.
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.
-
John Polidori. To pass the time, the group read a volume of ghost stories aloud, at which point Byron posed a challenge in which each member of the group would attempt to write such a tale.⏎“The chronology that’s in most
-
Baltimore and Washington, D.C., as well as in Florida and some Southern states.⏎Don’t Hebrew Israelites have a community in southern Israel?⏎One part of the wider Black Hebrew Israelite movement began arriving in Israel in the late 19
-
attributes. Each node in the tree represents a decision based on a feature, and the tree continues to split the
-
be fading away. english⏎and its speakers, owing to their advantage, are going to dominate societies. If members of the elite class in countries where english⏎is not the spoken language, live their lives
-
of the very best Old English/Anglo-Saxon poems is followed by footnotes and the translator's comments. Included in the⏎notes are a summary and a detailed analysis of the⏎poem's plot
-
house thousands of Japanese American internees. One of the first families to arrive was the Wakatsukis, who were ordered to leave
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 number of, and the number of, and the number of , ... for ... x millionth, which is how much it is for 10% of 24. So the answer is 10% of 2
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. In this article, we will look at five strategies for motivating students and keeping their brains engaged over the summer.⏎Engage in Reading Adventures⏎There is no better time to read than during the dog days of summer. Whether it’s sci-fi
-
3-million people (more than one-third of the entire U.S. population) to the joyous melodies and excitement of the Fab Four.⏎Nothing would ever be the same
-
had been handed down over the years by family and friends and both finally landed in the lap of Blight, who is director of Yale University's Gilder Lehrman Center for the Study of Slavery, Resistance, and Abolition.⏎Even
-
update on: 2006-SEP-23 Matthew 12:30, Luke 11:23, and Romans 16:17. Unfortunately, it is the other verses that are more frequently
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
- L6:44 suppresses
- L6:3 excites
- L6:187 suppresses blank token before a number or after a count
- L6:93 suppresses
- L5:30 suppresses
- L5:248 weakly suppresses
- L6:376 weakly suppresses
- L6:301 weakly suppresses
- L1:2157 weakly suppresses
- L6:226 weakly excites
- L3:2766 weakly suppresses
- L6:489 weakly excites
- L5:10 weakly excites
- L5:2156 weakly suppresses
- L6:2146 weakly excites
- L1:2866 weakly excites
- L2:224 weakly excites
- L2:411 weakly excites
- L6:10 weakly excites
- L1:194 weakly excites
- L2:1908 weakly excites
- L6:346 weakly suppresses
- L4:15 weakly excites
- L1:2700 weakly excites
Feeds later layers
- L8:1087 weakly excites
- L10:17 weakly excites
- L8:772 weakly suppresses "it" as a subject pronoun
- L17:3007 weakly excites
- L11:315 weakly suppresses
- L10:97 weakly excites
- L9:191 weakly suppresses
- L16:2655 weakly suppresses
- L10:1448 weakly excites
- L11:23 weakly excites
- L23:1234 weakly excites word pieces inside names and terms
- L11:378 weakly excites
- L20:379 weakly suppresses digits inside dates
- L12:2186 weakly suppresses
- L16:1006 weakly suppresses
- L22:1073 weakly suppresses
- L24:2308 weakly suppresses
- L16:527 weakly excites
- L19:1174 weakly suppresses
- L21:675 weakly suppresses
- L15:779 weakly suppresses
- L17:30 weakly excites
- L23:2964 weakly suppresses
- L8:315 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
- @brief
- ArrayOf
- Handling
- InOut
- 著
- bane
- -bl
- esch
- -li
- eros
Suppresses
- груз
- utta
- ата
- iden
- ��
- utory
- руз
- 意见反馈
- ottage
- thị