'into' after a verb or 'well'
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. 19 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 word 'into' when it follows a verb or phrases like 'well into', as in 'vanishes into the past'. Turning it up did not produce any clear theme in the writing.
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.
-
Until well into the 20th century, knowledge of Jacques Le Moyne de Morgues
-
4 “Glory to God in the highest,⏎and on earth peace among those with whom he is pleased!”⏎15 When the angels went away from them into heaven, the shepherds said to
-
which seems to have been written by their friend, Hugo Dyson! When they rush to warn him, Hugo walks through a door in time - and vanishes into the past. And in that moment, the world begins to change.The Frontier,
-
and warm, southern ones. You can enjoy it for three whole seasons every year with varieties which flower as early as March, and others which produce flowers until well into the autumn. This makes
-
oughly 180 million years ago, during the height of the Jurassic period, the Earth's magnetic field flipped, bringing the magnetic north pole once again into the Northern Hemisphere.⏎This so-called van Zijl reversal, named for the
-
movements for our freedom. He has faced many struggles went into prison for many times. After his great struggle, we got freedom on 1
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: that in this 500% (100000000000000000000000000000000000000
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. The American Academy of Child & Adolescent Psychiatry believes the problem is even more widespread, reporting that as many as 50 percent of all young people are bullied at least once during their school years, with about one in 10 suffering from
-
he shared:⏎The tragic Ghost Ship fire in Oakland has reminded us all that fire can kill very quickly. And it is not the only danger that we may face at home and in public places. A few basic precautions can save your life — but
-
nonetheless. Here are two important reasons why:⏎The odds of needing long-term care are high:⏎- Approximately 70% of people will need long-term care at some point during their lifetimes after reaching age 65.*⏎- Approximately
-
:⏎- Protected classes in California⏎- Anti-discrimination requirements for employers in California⏎- What to do if your protected class is discriminated against⏎Facing discrimination at your workplace? You came to the right place.⏎Protected Classes in California⏎In
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
- L10:132 excites
- L18:135 excites
- L16:825 weakly excites
- L4:1406 weakly excites
- L12:2522 weakly excites
- L8:1627 weakly excites
- L4:328 weakly suppresses 'to' after 'so as' and infinitive cues
- L13:2167 weakly excites
- L18:1069 weakly suppresses
- L11:293 weakly suppresses
- L6:1149 weakly suppresses
- L15:317 weakly excites
- L2:1579 weakly suppresses
- L12:559 weakly suppresses
- L4:2899 weakly suppresses
- L5:2992 weakly excites
- L3:2929 weakly excites
- L4:1847 weakly suppresses
- L15:1328 weakly suppresses
- L10:944 weakly suppresses
- L6:515 weakly suppresses
- L3:687 weakly excites
- L5:3009 weakly excites 'to' after a verb or noun
- L0:2383 weakly excites
Feeds later layers
- L21:1935 excites
- L22:2262 excites
- L25:2648 suppresses
- L20:1139 excites 'in' after a past participle
- L24:893 weakly suppresses
- L21:2823 weakly excites
- L24:2584 weakly excites 'in' in set phrases like 'in mind'
- L20:83 weakly excites
- L23:2125 weakly suppresses
- L20:1722 weakly excites
- L22:2998 weakly excites
- L21:2170 weakly excites
- L24:1442 weakly excites
- L25:3070 weakly excites
- L27:1851 weakly suppresses capital K starting a word
- L23:2136 weakly excites 'in' after a noun phrase
- L21:1081 weakly excites
- L25:1865 weakly excites 'in' in 'keep in mind'
- L24:1957 weakly excites capital 'In' starting a word or name
- L21:1335 weakly suppresses
- L24:2163 weakly excites
- L25:1853 weakly excites
- L23:769 weakly suppresses
- L21:2772 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
- liğinde
- osphere
- ther
- ther
- nowhere
- oto
- odon
- ducted
- _collections
- BuildContext
Suppresses
- hammad
- 섭
- 人と
- sti
- งาม
- root
- Ctl
- -notch
- 泪
- #undef