'in' after a motion verb
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. 16 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 'in' when it follows a verb of movement, such as 'going in', 'flew in', 'heaved in' or 'flows in'. Turned down, the model drifted into number talk. Turned up, it wrote unrelated sentences, so the steering did not clearly confirm 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.
-
worked, underfed horses. This action caused many streetcars to be stalled behind the first, now horseless, car. Bergh then stopped a car going in another direction.⏎The “man in the black hat” had been trying for many
-
Inside a sponge there is a system of openings and passages called a canal system. The sponge is nourished by chemicals that they filter out of the sea water passing in and out of these canals. Mature sponges do not move around.
-
more scarce because of human development, Dingoes are turning to domesticated animals for food, pitting themselves against the sheep farmers.⏎“Quite often they’ll go in and just kill a whole paddock full of sheep, and just leave it.
-
in the new park. It was chaos.⏎After much research, it was discovered that the absents of the mature bull elephants was the culprit. Quickly they flew in male role models and the bizarre and violent behavior of the juvenile elephants stopped completely in
-
a nice sheltered spot. It would even be safe from the occasional sudden great storms that struck without warning in that late summer season. As the men heaved in the anchor and the ship swung around to face the breeze, the master ordered several
-
Namibia, a former diamond mining town whose population steadily declined after WWI, when diamond prices crashed, and Balestrino, Italy, whose inhabitants were moved in 1953 after several earthquakes brought them to the conclusion that the area
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: the I: I 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.
-
221BC). Unlike most Chinese idioms, 毋忘在莒 strangely has actually two stories of origin, which gave this idiom two different
-
exists in the mind as a sensation and appears to us in various hues” (Morton
-
法、分析法来进行研究,探讨泰国学生在学习汉语助动词中的困难及偏误情况,分析他们产生偏
-
to be an accessible topic for middle school students.19<|endoftext|>毋忘在莒(Pinyin: Wu Wang Zai Ju) is a famous Chinese idiom
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
- L15:1328 suppresses
- L4:289 excites
- L9:1025 excites
- L14:1389 suppresses
- L3:493 suppresses
- L4:1966 suppresses
- L0:781 suppresses
- L10:262 suppresses
- L5:490 excites
- L1:1179 suppresses 'in' after 'result' (results in)
- L8:2438 suppresses
- L2:2307 excites
- L4:463 suppresses
- L8:351 suppresses 'in' after 'high' or 'dressed'
- L13:132 excites "of" after a noun
- L6:3070 suppresses
- L10:132 excites
- L4:2275 suppresses
- L3:1116 excites
- L0:1109 excites
- L14:1625 excites the word 'in' after a noun
- L6:846 weakly suppresses
- L4:16 weakly excites
- L4:1016 weakly excites
Feeds later layers
- L17:1434 strongly suppresses
- L17:2821 strongly suppresses
- L19:1345 strongly suppresses
- L18:1588 suppresses
- L17:165 suppresses
- L21:1269 suppresses "the" in "on the other hand"
- L27:1363 suppresses
- L19:1464 suppresses
- L25:1865 suppresses 'in' in 'keep in mind'
- L18:483 excites
- L23:1366 suppresses
- L17:2302 suppresses 'in' after a noun
- L17:2609 suppresses
- L20:2850 suppresses
- L23:2136 suppresses 'in' after a noun phrase
- L19:2775 suppresses
- L19:947 excites
- L21:3039 suppresses
- L19:1840 suppresses
- L24:1957 suppresses capital 'In' starting a word or name
- L18:2356 suppresses
- L19:2363 suppresses
- L17:2619 suppresses
- L20:2201 suppresses 'the' in location phrases like 'at the end'
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
- thái
- тель
- тельные
- ancement
- �单
- �
- мы
- �数
- ngại
- ��
Suppresses
- 屏幕上
- 一定程度
- 人群中
- 实践中
- {}]
- 范围内
- 活动中
- 不利
- vented
- 桌子上