closing parenthesis in data 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. 27 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
It fires very strongly on the closing bracket ')' at the end of a short note in a table or list, such as '(from vid)' or '(63% off)'. Turning it down made the model produce runs of quotation marks and punctuation, which fits a neuron that works against piling on more punctuation.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
/MSE) = 10 log10(MAX/(MSE)^(1/2))^2 = 20 log10(MAX/(MSE)^(1/2)). Therefore, PSNR = <|endoftext|>In my current position
-
|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|>
-
Date published: || 2015|⏎|ISBN: ||9781498738552 (DRM-PDF)|⏎|Read Aloud: ||not allowed|<|endoftext|>Under technical guidance and collaboration
-
Sanskrit.⏎|karsharpan / karsha (coin)||cash|⏎|mukh (face)||Mouth (old english muth)|⏎|kapal (head)||as in per “capita”, captain,
-
)⏎|Council of Trent|⏎|Accepted by||Catholic Church|⏎|Fifth Council of the Lateran (1512–1517)|⏎|First Vatican Council (1869–1870)
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:."."."."."."."."."."."."."."."."."."."."."."."."
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
extension of Bullseye matrix problem.⏎Who has power to do everything in this world?⏎Close MATLAB with keyboard⏎Find the treasures in MATLAB Central and discover how the community can help you!⏎Choose a web site to get translated content where available and
-
universal and regional human rights instruments such as the International Covenant on Civil and Political Rights. The African Charter, and the International Convention on Social Economical and
-
iblocs. What is the same and⏎what is different about these pairs of shapes? Can you describe the⏎shapes in the picture?⏎10 space travellers are waiting to board their spaceships. There⏎are two rows of seats in the
-
1, pp. 44-69. ISSN 1980-4369. http://dx.doi.org/10.1590/S0101-9
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
- L25:36 strongly suppresses
- L26:153 strongly suppresses
- L26:381 strongly suppresses
- L22:271 strongly suppresses
- L20:863 strongly excites
- L25:118 strongly excites
- L16:2 excites
- L26:1142 excites
- L21:50 excites comma after 'So' or intro phrase
- L20:344 excites hyphen inside compound words
- L18:48 suppresses
- L26:1314 excites
- L21:2374 excites
- L26:712 excites
- L23:35 excites
- L19:1777 suppresses
- L25:112 excites
- L23:109 excites
- L18:430 suppresses
- L26:1179 excites
- L23:77 excites the space right before a number
- L25:814 suppresses
- L24:1223 suppresses
- L26:345 excites
Feeds later layers
None: this is the last layer.
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
- ibern
- fäll
- ickle
- !”
- erne
- าม
- i
- посколь
- irl
- ,
Suppresses
- .
- "
- )
- -
- (
- *
- :
- /
- [
- ,