line breaks inside 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. 2 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 responds to the invisible line break at the end of an item in a list, just before the next item starts. It is most active in lists of short points or answer options. Turning it up did not produce a clear, matching change in the output.
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.
-
14)⏎- Healthy chicken recipes⏎- Meatless meals: The benefits of eating less meat⏎- Healthy cooking for 1 or 2⏎- see all in Healthy cooking
-
years at a time. Read more about lay judges.⏎Typical cases in the district court⏎Cases in the district court are divided into three categories; contentious cases, criminal cases
-
of the uterus, the lining of the uterus⏎Severe stomach inflammation⏎Liver and kidney toxicity⏎Changes in the level of fats in blood and signs of liver⏎There’s A LOT of evidence out there and it’s not quoted by the bi
-
allergic reactions⏎Enlarged lymph nodes and immune disturbances⏎Disturbed liver, pancreas, and testes function⏎Disturbed enzyme functioning in kidney and heart⏎Higher
-
B. What is the most appropriate treatment?⏎MRI of the tibia⏎Aspiration of the tibia⏎Referral to child services⏎Long leg cast application⏎Serum vitamin D, calcium, and phosphate levels⏎Select Answer to see Preferred
-
etic techniques that can be used⏎Sterilize surfaces and working surfaces⏎Washing hand before and after⏎Flaming inoculating loop and lip of culture tubes⏎Reduce time sterile medium, cultures, or bacteria are exposed to air⏎Work in
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 Am A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A, A,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
into Germany.⏎This led to alleged rifts within the German squad and distress among some spectators during Germany’s games.⏎History of the world cups⏎How the football World Cup finally came about in 1930⏎The controversies, shocks at
-
what it ate. But perhaps they knew that jumping spiders were cat-like in their movements, and [they] decided to name the spider after the agile panther Bagheera in Kipling's book."⏎"Utterly Surreal"
-
data represent the age Of U.S. presidents on their respective inauguration days(through Barack Obama). (a
-
⏎(f) Louis Daguerre Typewriter⏎- Name the inventors of the following:⏎(b) ball-point pen⏎- Elias Howe made the first sewing Machine. Who improved it?⏎- Who invented the Computer?⏎- Who is credited with
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
- L0:212 strongly suppresses
- L1:77 strongly suppresses
- L0:300 strongly suppresses
- L1:893 strongly suppresses
- L1:574 strongly excites
- L0:51 excites
- L1:762 excites
- L1:655 excites
- L0:138 excites
- L1:1221 excites
- L1:116 excites
- L1:634 excites
- L1:71 suppresses
- L0:178 suppresses
- L0:1224 suppresses
- L0:1301 suppresses
- L0:148 suppresses
- L0:115 excites
- L1:940 excites
- L0:351 weakly excites
- L1:824 weakly excites
- L0:2229 weakly excites
- L0:306 weakly excites
- L0:1904 weakly suppresses
Feeds later layers
- L5:114 strongly suppresses
- L3:21 strongly suppresses
- L3:195 suppresses
- L5:1286 suppresses
- L3:1057 suppresses
- L4:176 suppresses
- L6:173 excites
- L7:1302 suppresses
- L5:212 excites
- L10:1274 excites
- L5:79 suppresses
- L6:189 suppresses
- L10:471 suppresses
- L15:1395 excites
- L16:847 suppresses
- L4:495 excites
- L3:169 excites
- L4:30 excites
- L23:242 excites
- L15:790 suppresses
- L9:235 suppresses
- L3:45 excites
- L20:296 excites
- L21:370 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
- :expr
- :index
- .preprocessing
- 惯例
- (void
- 莒
- IMS
- 均衡
- 扩张
- bất
Suppresses
- fully
- urve
- ariant
- ductive
- .Now
- .Areas
- entialAction
- setFlash
- 蝇
- ochen