mid-sentence line breaks
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. 21 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 line-break characters that appear in the middle of a sentence, where the text was wrapped onto a new line. Turning it on made the model's text drift into unrelated topics, so the steering does not clearly show the 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.
-
else, themselves included.⏎Census-takers designated them Russian-born, and they were, literally.⏎Geographically they were as well, since they originated from at⏎least six distinct nationalities within the vast Russian Empire.⏎In every other respect,
-
CD.⏎High quality scanned images of the whole of the original book. This CD has been bookmarked for easy⏎navigation, and pages can be searched, browsed, enlarged and printed out if
-
uffed at the knees, a style that⏎involves folding the garment in half and tucking it under the waist. This style⏎of draping the⏎dhoti is known as the half-fold. The dhoti can be
-
âyâna) is most commonly known in the Chinese Buddhist literature by the name of Açvaghosha. But according⏎to his Life he was also called Kung-tê-jih (i.e
-
from organic matter as opposed to fossil fuels.⏎“As a renewable source of natural gas, the city’s sewage will offer a steady and sustainable supply of clean fuel well⏎into the future,” said Michael Jordan, Environmental Services’ director. “The filling
-
teaser just like this one...⏎Or, just get a random brain teaser⏎If you become a registered user you can vote on this brain teaser, keep track of⏎which ones you have seen, and even make your own.⏎Back to Top<|endoftext|>
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 will see, as I will see, the shape of the shape. To see the shape of the shape. That will mean, I think, that the shape is the meaning, to be the shape, to be the meaning. that
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
How do land use practices and management efforts interact with natural disturbances and climatological cycles to affect estuarine water quality and suitability for ecologically (and economically) important species?<|endoftext|>Not Just for Kids⏎The
-
to increased pressure inside the skull, including:⏎Brain herniation can occur:⏎Patients with a brain herniation have:⏎A brain and nervous system (neurological) exam shows changes in alertness (consciousness). Depending on the severity of the
-
ology Course Description⏎- 2 Case Studies⏎- 3 Outbreak Investigations⏎- 4 Clinical Trials⏎- 5 Cohort (Longitudinal) Studies⏎- 6 Case-Control Studies⏎- 7 Mesures of Mor
-
parallel to the sun (for example, sunrise striking a horizontal rooftop), the angle of incidence is 90°.⏎Array — see photovoltaic (PV) array.⏎Battery Cell — The simplest operating unit in a storage battery. It consists
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
- L20:564 strongly suppresses
- L20:1147 strongly suppresses comma after a sentence subject
- L20:281 strongly excites
- L16:2076 strongly suppresses
- L20:1447 strongly suppresses
- L16:1323 suppresses
- L16:1258 suppresses
- L17:310 suppresses
- L16:2315 excites
- L16:2983 excites
- L18:933 suppresses
- L20:1595 suppresses
- L20:1781 suppresses
- L20:526 suppresses
- L20:90 excites
- L9:1515 excites line break splitting a phrase
- L17:1192 suppresses
- L16:1003 excites
- L16:1608 suppresses
- L18:237 suppresses
- L15:2449 excites line breaks mid-sentence
- L17:359 suppresses
- L18:879 excites
- L16:1246 suppresses
Feeds later layers
- L27:3020 weakly excites
- L26:2128 weakly excites decimal points inside numbers
- L26:2731 weakly excites
- L27:2135 weakly excites
- L26:747 weakly excites
- L23:1624 weakly suppresses
- L25:1279 weakly excites
- L26:148 weakly excites
- L24:662 weakly suppresses
- L25:387 weakly excites
- L27:362 weakly excites
- L27:2258 weakly excites
- L22:2089 weakly suppresses
- L23:1185 weakly excites
- L27:1354 weakly excites
- L25:645 weakly excites
- L23:334 weakly excites
- L27:369 weakly excites
- L25:969 weakly excites
- L25:1736 weakly excites
- L23:87 weakly excites
- L26:2634 weakly excites
- L25:808 weakly excites
- L25:2869 weakly excites
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
- rish
- же
- справ
- łu
- ện
- ὕ
- ích
- 下来
- �
- edio
Suppresses
- /etc
- AFC
- alike
- Traditional
- ロン
- imulation
- conversion
- 틱
- Alicia
- common