colon that introduces a statement
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. 14 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 reacts to a colon that is about to introduce an explanation, quote or example, such as "I start to wonder:" or "here's one example:". Turning it down made the model output strings of punctuation, and turning it up gave rambling 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.
-
we have almost no idea about how powerful the military technologies of the aliens can be and whether they will be friendly to the human or not; I start to wonder: ‘Assuming there is only finite resource on the earth based on which we cannot
-
, many literature professors have taken an extreme position that authorial intent should never be considered, ever. To even look down that road is foolish.⏎But consider this: if you walk into any contemporary art museum you will, without fail, find on
-
the interruption would last. But with her intimate knowledge of volcanic ash, and a good idea of what it does to jet engines, she was sure of one thing: “I wouldn’t want to get on a plane right now.”⏎Admirably
-
benefits to agriculture.⏎For more information, please contact:⏎Dr. Norihito Nakamichi⏎Biodynamics Research Team⏎RIKEN Plant Science Center⏎Tel: +81
-
, Comparative Physiology also gave birth to an experimental approach that has come to be known as the “August Krogh Principle,” which states: “For many problems, there is an animal on which it can be most conveniently
-
time.⏎You should give examples from dynamic typed languages instead, e.g. clojure.⏎Here's one example:<|endoftext|>The commerce of the ancient world was gathered into the warehouses of Tyre.
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::::::::::::::::::::::::: : ::::::::::: :: is that??:? is this for:
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Why Lesson Planet?⏎Find quality lesson planning resources, fast!⏎Share & remix collections to collaborate.⏎Organize your curriculum with collections. Easy!⏎
-
poverty group. A family from this group with four members, which is the average family size in the<|endoftext|>Why Lesson Planet?⏎Find quality lesson planning resources, fast!⏎Share & remix collections to collaborate.⏎Organize your curriculum with collections. Easy!⏎
-
test, hemogram, ESR⏎with antibiotics and anti-inflammatory medications helps reduce tissue damage⏎from inflammation, lessen pain
-
?, on Time's website.<|endoftext|>Start a 10-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!⏎
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
- L9:7 strongly excites
- L1:1 strongly excites
- L10:33 strongly suppresses
- L12:5 strongly excites
- L2:204 suppresses
- L3:3 suppresses
- L6:4 suppresses
- L8:2 excites
- L1:8 excites
- L6:0 excites
- L2:44 excites
- L5:5 suppresses digit after a label like paper or type
- L13:10 weakly excites
- L0:1224 weakly suppresses
- L8:30 weakly excites blank slot right before a number
- L10:86 weakly suppresses
- L0:1301 weakly suppresses
- L13:345 weakly excites
- L0:178 weakly suppresses
- L0:1864 weakly excites
- L13:95 weakly excites
- L1:106 weakly excites
- L9:2517 weakly suppresses
- L3:2331 weakly suppresses
Feeds later layers
- L15:123 strongly excites
- L18:48 strongly excites
- L17:38 strongly excites
- L16:170 strongly excites
- L15:44 strongly suppresses
- L21:173 strongly excites
- L20:863 strongly excites
- L20:8 strongly suppresses
- L17:62 strongly suppresses
- L19:131 strongly suppresses
- L25:118 strongly excites
- L23:35 strongly excites
- L22:262 strongly excites comma inside big numbers
- L21:195 strongly excites
- L17:137 strongly suppresses
- L16:241 strongly suppresses
- L25:36 excites
- L18:887 excites
- L27:143 excites
- L26:94 excites
- L18:246 excites
- L25:273 excites dollar sign before an amount
- L24:130 excites line break at the end of a line
- L20:836 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
- 特别是
- anca
- 尤其是
- 特别是在
- merce
- idents
- rawler
- 榔
- ([]*
- 一门
Suppresses
- :
- (Y
- :y
- -packed
- :t
- :E
- (y
- (d
- =d
- (t