the first dot of an ellipsis
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. 26 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 a full stop that is starting a trailing dot sequence like ". . .", often in quoted speech or a sentence that trails off. When it was turned down, the model got stuck writing long rows of dots.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
speed in chatrooms and on social networking sites where the general attitude is that there isn’t a need . . . to conform to spelling rules,” the study stated.⏎“We are now witnessing the
-
it as a lone piece of evidence:⏎One of the great problems of evolution has been to find anatomical links between the fishes and their land-invading descendants . . . For a long time evolutionists were troubled by this major gap between fishes and
-
lizards!⏎Ecologist Sean Doody discovered this while in the field with student Phillip Paull. The two were lifting eggs out of a crevice when . . . Holy Dooley! The hatchlings literally popped out and sprinted away!⏎
-
the surface with thin layers of pigment, the color of the underlying layers modifying those on top. Painter Gino Severini reported that "Matisse said . . . that everything that did not contribute to the balance and rhythm of [this]
-
50 to 60 percent mercury. The filling material is then placed in the tooth and smoothed with various dental tools.⏎|DID YOU KNOW . . .⏎Mercury (Hg), sometimes known as quicksilver, is named for
-
; and thence, by that parallel of latitude, to the South Sea." In addition, the treaty stated that "all islands in the Sabine, . . . Red and Arkansas Rivers . . . [were] to belong to the United
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.
-
πον ἐγένετο καὶ οὐχ ὁ ἄνθρωπος διὰ τὸ σάββατον·⏎2 M
-
’ οὐ διὰ τὸν τόπον τὸ ἔθνος⏎ἀλλὰ διὰ τὸ ἔθνος τὸν τό
-
0-Day Free Trial to Unlock the Full Review⏎Why Lesson Planet?⏎Find quality lesson planning resources, fast!⏎Share & remix collections to collaborate.⏎Organize your curriculum with collections. Easy!⏎Have time to be more creative & energetic with your students
-
Rapid City, fire weather meteorologists are watching conditions closely. And until we receive widespread, heavy moisture they'll be monitoring what is known as the Keetch-Byram Drought Index or KBDI.⏎It measures the amount of precipitation needed
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
- L23:1580 excites token before a statistic or number
- L25:1326 excites ellipsis dots after a cut-off quote
- L25:112 excites
- L23:372 suppresses
- L23:1196 suppresses
- L25:725 excites
- L25:1090 suppresses
- L22:2565 suppresses
- L24:1960 excites
- L23:1789 excites
- L23:480 excites
- L24:2475 suppresses
- L21:1405 suppresses
- L25:981 weakly excites
- L23:777 weakly excites
- L24:1539 weakly excites
- L23:508 weakly suppresses
- L19:679 weakly suppresses
- L8:12 weakly excites
- L25:294 weakly excites
- L1:117 weakly suppresses
- L19:1516 weakly suppresses
- L8:14 weakly excites
- L22:1963 weakly suppresses
Feeds later layers
- L27:319 strongly suppresses
- L27:25 strongly excites
- L27:385 suppresses
- L27:703 suppresses
- L27:1418 suppresses
- L27:731 excites
- L27:366 suppresses
- L27:816 weakly suppresses
- L27:362 weakly suppresses
- L27:1328 weakly suppresses
- L27:574 weakly suppresses 'Aut' starts and colon prediction
- L27:764 weakly excites
- L27:426 weakly suppresses
- L27:1045 weakly suppresses
- L27:370 weakly suppresses
- L27:395 weakly suppresses
- L27:701 weakly suppresses
- L27:569 weakly suppresses
- L27:351 weakly suppresses
- L27:669 weakly suppresses
- L27:977 weakly suppresses
- L27:782 weakly suppresses
- L27:1120 weakly suppresses
- L27:77 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
- iners
- çı
- -going
- patible
- ulary
- лё
- attro
- rats
- ycling
- META
Suppresses
- .").
- .).
- .</
- .\"
- .).⏎⏎
- .--
- .`
- .', ⏎
- ."); ⏎
- .';⏎