closing parentheses in lists
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. 24 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 closing bracket that ends a short note in a list or table, like the ")" after a measurement, a year range, or a format name. Changing it did not produce a clear theme in the generated text.
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.
-
|Length:||205 ft (62.48 m)o/a|⏎|Beam:||33 ft (10.06 m)|⏎|Draught:||11.5 ft (3.5
-
, sugar|⏎|vak, vaacha||voice, vocal|⏎|veda, vidya (from vid) (knowledge)||wit, wise (from vid)|⏎The postfix ‘er’ in words indicates the occupation / action of a<|endoftext|>
-
|miti (measure)||metric, meter, measure (latin metiri)|⏎|gouh (earth)||geometry, geography (greek geos)|⏎|lok (place)||location, locus, locale|⏎|ek (one
-
ategorize this paper)|⏎|Buy the book||$8.32 used (63% off) $17.25 new (22% off) $21
-
' Classroom.|⏎|Levels:||Late Elem. (3-5), Middle School (6-8), High School (9-12)|⏎|Resource Types:||Public and Government Institutions, Web Interactive/Java|⏎
-
nose, nostril, nasal|⏎|danta (tooth)||dental, denture|⏎|janu||knee (french genou)|⏎|pad (feet)||pedicure, pedestrian, centipede
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 goal is to make the world a home for all people.\ \ Now it seems that some people might justifiar'tc that the goal is a bad option because some people might not be able to self-resource themselves (they
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
) testifies (Adv. Har. xxix.1) It is derived from the Hebrew Ebion, "poor," and was not given, as Origen supposes, for their low view of Christ. Id. at 6
-
. However, right around 235, Hippolytus was reconciled with the Church.⏎One of the other items on Pontian’s plate was Origen. As a young man, Origen established a school of philosophy in Alexandria
-
37, has been selected as an Associate Laureate in the Rolex Awards for an ambitious project to encourage more than 500 families in the areas around Orurillo and Pucará to broaden the genetic variety of their crops. More
-
download these videos from Media Hopper Create.⏎[Featured image: US Flag Backlit, CC BY SA, Jnn13, Wikimedia Commons]<|endoftext|>Coronary artery
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
- L19:118 strongly suppresses
- L0:304 strongly suppresses
- L16:39 strongly excites
- L16:97 excites
- L16:0 suppresses
- L4:489 suppresses
- L19:1636 excites closing bracket after a list number or letter
- L16:13 suppresses
- L22:866 suppresses
- L16:9 excites
- L16:61 excites
- L20:61 suppresses
- L18:569 excites
- L16:2 suppresses
- L2:1808 excites
- L19:233 excites
- L20:441 suppresses
- L2:1834 excites
- L23:1167 excites
- L21:173 excites
- L16:17 excites
- L16:94 suppresses
- L2:75 suppresses
- L4:412 suppresses
Feeds later layers
- L26:362 strongly excites
- L26:381 strongly excites
- L26:3057 excites
- L27:1305 suppresses
- L27:1687 suppresses
- L27:2174 suppresses
- L26:725 suppresses
- L27:2250 suppresses
- L26:236 weakly suppresses
- L27:429 weakly excites
- L25:1312 weakly suppresses unclear: mixed, Chinese and name fragments
- L27:613 weakly suppresses
- L25:483 weakly excites
- L26:1864 weakly suppresses
- L27:2714 weakly suppresses
- L25:422 weakly suppresses
- L27:657 weakly excites
- L27:1147 weakly suppresses
- L27:139 weakly excites
- L27:145 weakly suppresses dots inside web addresses
- L26:1138 weakly suppresses
- L25:244 weakly excites
- L26:193 weakly suppresses
- L27:1113 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
- ribbon
- iscrim
- isce
- cluded
- ris
- clusion
- vol
- 看书
- onium
- atten
Suppresses
- ,
- ,在
- cq
- ,并
- ،
- ,《
- //
- cj
- tement
- statt