'in' after motion verbs
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 fires on the word "in" when it follows verbs of moving or pushing, like "move in packs" or "pushing in all directions". Turning it up did not give a clear, consistent change in the model's 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.
-
ales are usually smaller. They are generally about 5-7 feet long from nose to tail (the tail is usually about 2 feet long) and move in packs of up to 15 members! To learn more about gray wolves,
-
tall cone such as Liberty Cap increases in height, the flow of water from the top may eventually stop, because there is
-
production of plastic materials has steadily increased in recent years. Because of their versatile and resilient nature, plastics play a critical role
-
forward direction - that is, it thrusts the airplane forward.⏎Jet propulsion is similar to the release of an inflated balloon. The pressure inside the balloon is pushing in all directions. It is also "jetting" out from the mouth of the
-
international, dry bulk goods use bulk ships. Liquid bulk moves in ocean-going tankers. General cargoes move in container ships or break bulk ships
-
0 minutes – it’s essentially moving in a straight line. The reason the path is all bendy is because Swift itself
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 working on this. I have been working on this. I have been working on this. I have been working on this. I have been working on this. I have been working on this. I have been working on this
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
“white” turkeys found in North America are actually Broad-breasted Whites, which are a cross between White Holland
-
. This can also cause dehydration and vitamin and mineral deficiencies. Treatment may require hospitalization.⏎HG is a more severe form of nausea and vomiting of pregnancy (NVP), also called morning sickness. Morning sickness affects anywhere between 50% to
-
their ability to meet and worship, and has helped people in prisons, jails, mental institutions, and state-run nursing homes preserve their religious freedom.⏎The values embodied in RLUIPA are universal ideals. Department of Justice attorneys have provided technical assistance on
-
of inaccuracies and misfires.⏎Truth, when hyped, loses credibility. There have been attempts, of late, to eulogise ancient theories contained in ayurvedic classics as ultra-modern ‘quantum physiology’1.
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
- L0:38 weakly suppresses
- L0:781 weakly suppresses
- L1:2413 weakly excites
- L1:1110 weakly excites
- L0:880 weakly excites
- L0:2641 weakly suppresses
- L0:1311 weakly suppresses
- L0:2162 weakly suppresses
- L0:1316 weakly suppresses
- L0:1405 weakly excites
- L1:1724 weakly excites
- L1:2904 weakly suppresses
- L0:204 weakly suppresses
- L1:1131 weakly suppresses
- L1:2859 weakly excites
- L1:2722 weakly suppresses
- L0:2374 weakly excites
- L1:1346 weakly excites
- L0:624 weakly suppresses
- L0:1224 weakly suppresses
- L1:2042 weakly excites
- L0:178 weakly suppresses
- L1:1 weakly suppresses
- L0:2951 weakly excites
Feeds later layers
- L3:2778 weakly suppresses
- L13:2527 weakly suppresses
- L14:1939 weakly suppresses
- L5:2204 weakly excites
- L13:1213 weakly suppresses
- L7:2562 weakly suppresses
- L5:2849 weakly excites
- L13:383 weakly suppresses
- L6:1997 weakly excites
- L13:1921 weakly excites
- L16:2903 weakly suppresses
- L9:363 weakly excites
- L10:2908 weakly excites
- L12:544 weakly suppresses
- L3:184 weakly suppresses
- L16:607 weakly suppresses
- L12:1870 weakly suppresses
- L8:932 weakly suppresses
- L19:2494 weakly suppresses
- L11:2176 weakly suppresses
- L11:123 weakly suppresses
- L24:2849 weakly suppresses
- L9:2521 weakly suppresses
- L15:336 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
- '~
- imeType
- pecies
- order
- leness
- ưỡ
- olars
- ited
- :type
- trình
Suppresses
- gage
- unted
- stra
- ANCH
- doubted
- ATEST
- amento
- .stem
- _CAST
- aten