opening bracket after a term or number
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 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 an opening bracket that comes right after a name, abbreviation, or measurement, where the text is about to give the short form or the converted value, like '2 kg (4.4 lb)'. Turning it up led to repetitive or off-topic text, which does not tell us much.
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.
-
. The U.S. military was also developing miniature UAVs that could be launched and operated by one person. The Marine Corps began testing a 2-kg (4.4-l
-
God) and the charming Temple of Baalshamin (god of fertility) — and a monumental arch that opens onto a 1,200-meter (3,937 ft.) long colonnade, the center of this
-
and study noctilucent - or 'night shining' - clouds, odd collections of ice crystals that form near the edge of space some 50 miles (80 kilometers) above Earth's polar regions. The clouds form so high above
-
that was more or less inevitable. It bothers me that people are so stupid as to use this stuff .”⏎By contrast, synthetic tetrahydrocannabinol (THC) is available as dronabinol (Marinol) and
-
Zionist journalism.⏎Already a member? Login here.<|endoftext|>The fractured rock surfaces were observed under different loading stages in creep shear by making use of computerized tomography (CT). Through the creep shear tests, the failure modes of the fractured rocks have
-
and the belief that we can recover from setbacks and learn from them.⏎Singaporean students have typically scored very well in the Organisation for Economic Co-operation and Development’s (OECD) Programme for International Student Assessment (PISA) in Reading, Mathematics
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: this is the time to think about the place of the place (the place is not here and this is the place). So these are these things are not the place (so those are the place). So these are these things are not the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
said that the image of Marx has been imprinted in his heart since he was a child, adding that “Marx molded by a Chinese artist contains emotion and [warmth].”<|endoftext|>Citadel of Herat⏎This citadel, which
-
biologist at the University of Arizona in Tucson.⏎"They had no idea what it ate. But perhaps they knew that jumping spiders were cat-like in their movements, and [they] decided to name the spider after the agile panther Bagheera in
-
assigned to eastern Chad to provide guidence to a literacy program.⏎“The reason that I’m back in Africa is because I left a piece of my heart here and [now] I’m here to get it back,” Eunice Kua said
-
which is really useful for us, because if you have the animals behave in the same way all the time, there’s no room for flexibility in the face of [climate] change.”⏎Even with a colony containing hundreds of
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
- L10:1459 suppresses
- L9:1065 suppresses
- L9:2010 suppresses
- L12:1443 excites opening parenthesis after a name or term
- L11:2774 weakly excites 'least' in 'at least' hedges
- L1:964 weakly excites "of" in "some kind of" or "some type of"
- L13:2302 weakly suppresses
- L1:1178 weakly excites
- L11:228 weakly excites
- L8:354 weakly excites opening parenthesis before a citation year
- L10:1643 weakly suppresses
- L6:990 weakly excites
- L0:1846 weakly excites
- L11:1077 weakly excites
- L2:981 weakly excites
- L10:1484 weakly suppresses
- L10:271 weakly excites
- L9:1421 weakly suppresses
- L5:330 weakly excites opening parenthesis after a noun
- L7:1590 weakly suppresses
- L13:553 weakly suppresses
- L8:2127 weakly excites
- L11:356 weakly suppresses
- L9:1899 weakly suppresses
Feeds later layers
- L15:1501 excites "and" joining two parts of a name or phrase
- L15:3004 excites
- L15:2831 excites
- L15:2437 excites opening parenthesis after a noun
- L15:1162 excites
- L15:734 excites
- L17:1679 excites
- L17:1214 excites
- L16:2049 excites
- L16:2201 excites
- L16:2289 excites 'or' linking two alternatives
- L16:882 excites
- L18:1386 excites unclear: words before 'and' or 'or'
- L15:232 excites
- L15:2215 weakly excites
- L15:3041 weakly excites
- L18:2136 weakly excites
- L17:2017 weakly excites
- L15:2884 weakly excites
- L16:247 weakly excites
- L15:2143 weakly excites
- L15:1380 weakly excites
- L15:1718 weakly suppresses
- L16:1074 weakly excites
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
- rogen
- ,
- nearly
- frame
- rarian
- REW
- resp
- /
- чив
- 党组书记
Suppresses
- 为中国
- initely
- |[
- Ǻ
- 时代中国
- istically
- [B
- ⽤
- locals
- Bahamas