the word 'in' after a verb or noun
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. 21 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 lights up on the small word 'in' when it starts a phrase that follows a verb or noun, like 'in order to' or 'in common with'. Turning it up made the model's text drift into unrelated topics such as grids, numbers and weather, so the link to the word 'in' is the only clear part.
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.
-
classroom in a setting where theory meets praxis. This course engages students in service-learning, which allows students to contribute their time and energy, in the context of academic coursework that involves reflection and intentional learning goals, in order to
-
sandstone.⏎[SinglePic not found]Every year thousands of local and international tourists visit the Phnom Da, Takeo. So many tourists visit the place in order to do a research on the well-known work of art of Cambodian
-
osis, I keep going round in circles!⏎Ok, can somebody please answer the following questions:⏎1. In me
-
being re-written. Until the TV movie based on Wells' novel was broadcast a few months later, this claim stood as being true. But once "The First Men In The Moon" aired, then Professor Cavor had to be credited with being
-
itself or in its effect”.⏎The constitutional anti-discrimination clause is set out in Article 23(2) that reads “no person shall be treated in a discriminatory manner by any person acting by virtue of any written law or in the
-
at the end of a person’s full name, after any given names.⏎What exactly is a surname?⏎Also called last name . the name that a person has in common with other
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 a 50% probability of not having a family n't any family n't any family n't any family n't any family n't any family n't any... n't any... n't any... n't any
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
are the primary kinds of antique lamps most favored by collectors. In the late 18th century, Swiss chemist Aime Argand invented the first lamp didn't require a free
-
the dark and the white races. We find the same differences in proportion between blondes and brunettes in the white races.⏎HAVELOCK ELLIS ON BLONDES AND BRUNETTES⏎The noted anthropologist Havelock Ellis says,
-
flak, flack, blast (intense adverse criticism) "Clinton directed his fire at the Republican Party"; "the government has come under attack"; "don't give me any flak"⏎- S: (v) open fire,
-
. Celery chipmunks and squirrels endangered, vulnerable, and threatened species flower gardening african wild dogs animal life zoos cats (felines) middle ages donkeys and mules papaya ducks domestic. This makes celery a good and healthy
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
- L19:595 strongly excites
- L18:282 strongly excites
- L3:410 strongly suppresses
- L16:55 strongly excites
- L10:1745 strongly excites
- L1:27 strongly excites
- L0:210 strongly excites
- L19:231 excites
- L17:391 excites
- L2:1923 suppresses
- L18:1279 excites
- L20:1040 excites
- L5:499 excites
- L0:24 suppresses
- L17:725 excites
- L14:95 excites
- L20:1659 suppresses
- L15:115 excites
- L15:96 suppresses
- L0:781 suppresses
- L20:2203 excites
- L20:435 excites
- L19:1639 suppresses
- L19:359 excites
Feeds later layers
- L26:2511 strongly suppresses
- L22:2299 strongly suppresses
- L23:1532 strongly suppresses
- L23:2762 strongly suppresses
- L27:3060 suppresses 'n' inside word fragments
- L24:2458 suppresses
- L22:2029 suppresses
- L25:2713 suppresses
- L25:890 suppresses
- L23:3041 suppresses
- L24:2929 suppresses
- L26:548 excites
- L27:2933 suppresses the letter n opening a word
- L27:2141 suppresses
- L26:2533 suppresses
- L25:2112 suppresses
- L22:1837 excites
- L27:2954 suppresses
- L23:2487 suppresses
- L24:2480 suppresses
- L23:1489 weakly excites
- L26:1408 weakly suppresses the 'n' that starts nectar
- L24:3047 weakly suppresses
- L27:2849 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
- din
- d
- tas
- eenth
- ennes
- tti
- irst
- tok
- een
- tif
Suppresses
- ners
- ning
- NING
- etwork
- ny
- nable
- ned
- ningen
- nen
- ights