digits inside long numbers
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 mixed: they peak on different tokens. 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 lights up on a single digit in the middle of a long number, such as a zero in an IP address or a digit in a citation number. When it was turned up, the model's writing fell into loops of repeated words, with no clear topic.
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.
-
6.0.0/15 IP: 10101100.00010000.00000000.00000000 SM:
-
heat pulse will switch the magnetic orientation of a nanometer-sized magnetic grain in just a few picoseconds (1 picosecond is 0.000000000001 seconds). As a comparison, today’s
-
fuel (£249,018,638).⏎- The estimated CO2 emissions from these miles would be 442,126.
-
33.0-38.4% of standard length; jaw teeth with slender shafts and broad crowns (Ref. 81260). Mature fishes of both sexes with variable black areas on lower jaw and lower
-
closely agree with your voltmeter readings.⏎Schematic with SPICE node numbers:⏎The two large-value resistors Rbogus1 and Rbogus1 are connected across the capacitors to provide a DC path to ground in order
-
1260), a tributary of the Kake River (affluent of Lake Barombi Mbo) (Ref. 81260) and Kumba stream (tributary of the
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 have been thinking about the idea of the first person, and I think about it from the beginning, and I think about it from the beginning, and, and, and, and, and, and, and, and, and,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Chemistry with his wife Irène Joliot-Curie for the discovery of artificial radioactivity.⏎|This scientist article is a stub. You can help Wikiquote by expanding it.|⏎- It is because I know all that science can bring
-
in Fundamentals of Epidemiology, a Johns Hopkins Bloomberg School of Public Health course with content available via JHSPH OpenCourseWare.⏎- 1 Fundamentals in Epidemiology Course Description⏎- 2 Case
-
⏎• Luke 9:58 Commentaries⏎• Luke 9:58 Bible Apps⏎• Luke 9:58 Biblia Paralela⏎• Luke 9:58 Chinese Bible⏎• Luke 9:
-
scanning (SEM) or transmission electron microscopy (TEM) coupled with energy dispersive x-ray spectroscopy (EDX).
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
- L10:856 suppresses
- L12:1902 excites
- L6:189 suppresses
- L10:390 suppresses
- L10:383 excites
- L10:3033 excites
- L4:687 suppresses
- L5:79 excites
- L11:2120 suppresses
- L6:58 excites
- L10:1837 excites
- L10:2084 suppresses
- L11:453 excites
- L9:708 excites
- L11:359 suppresses
- L10:134 excites
- L3:195 excites
- L13:695 excites
- L9:897 excites
- L4:495 suppresses
- L3:28 excites
- L4:103 excites comma after 'still' or 'in addition'
- L5:33 excites
- L11:325 suppresses
Feeds later layers
- L16:1872 strongly excites
- L15:1022 excites
- L15:890 suppresses
- L17:597 excites
- L19:1860 excites
- L15:1203 excites
- L15:658 suppresses
- L15:162 excites
- L15:488 suppresses
- L15:1160 excites
- L17:603 excites inside organisation and product names
- L18:595 excites
- L16:627 excites
- L15:486 excites
- L16:1249 excites
- L20:494 excites
- L20:225 excites
- L16:665 weakly suppresses
- L18:635 weakly suppresses
- L24:1604 weakly suppresses
- L17:685 weakly suppresses
- L16:1206 weakly excites
- L15:136 weakly suppresses
- L19:391 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
- _customize
- CardBody
- /version
- 的精神
- enario
- 蚌
- ponder
- llum
- eği
- 不信
Suppresses
- RF
- RH
- рож
- IJ
- R
- OSH
- FH
- ush
- RF
- aney