opening parenthesis after a citation or term
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. 10 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 an opening bracket that comes right after a name, citation, or term, where the text is about to add an explanation, a year, or a translation. Turning it up made the model drift into repetitive, off-topic rambling, which doesn't say much about the neuron itself.
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.
-
ak Dun v. Yukon, 2015 YKCA 18 (the “appeal decision”), the Yukon Court of Appeal unanimously allowed the Yuk
-
innovative drugs have been introduced, from Pfizer's (<|endoftext|>Information identified as archived on the Web is for reference, research or recordkeeping
-
the use of English and official language in the content classrooms, typically mathematics and science. According to Cummins, bilingual education is defined as “the used of two (or more) language of instruction at some point in a student’s school career”
-
muscle group is necessary for optimal muscle growth. For instance, Fonseca et al. (2014) showed that combining different lower-body exercises resulted in muscle growth
-
, 168 note) who remarks: — "Um dies (i.e. Varro's description of the extent of Roma quadrata) mit
-
“ponte las pilas,” (get to work,) are the speak of Latino Pachucos of the
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 know the words for it, and I know the meaning. And I know it's a very simple concept. But how to think about it? I know that it's a very simple concept. It is the most basic level of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
send, (and hopefully, receive), a valentine.⏎In the 1840s
-
innovative ways to diagnose the disorder earlier (as we reported here, here and here) and on new genes that contribute to it (which we described here).⏎(More on TIME.com: Health Checkup: Kids and Mental Health)⏎For the new
-
with opportunities for continued learning and for future employment, both domestically and internationally, in many areas such as commerce,<|endoftext|>Ancient stone circle holds air of mystery (it's in Britain, but it's not Stonehenge)⏎However, don't
-
marks the celebration the birth of Jesus Christ. Families usually gather on Christmas Eve or Christmas day to give presents. Households are decorated with lights and Christmas trees (and much more).⏎This brief synopsis illustrates that the major holidays of various traditions share a
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
- L4:64 strongly excites
- L7:100 strongly suppresses
- L9:982 strongly suppresses
- L4:45 strongly suppresses
- L6:468 strongly excites
- L6:409 excites
- L0:11 excites
- L1:155 suppresses
- L5:59 suppresses
- L9:886 suppresses
- L4:409 excites
- L7:732 excites
- L5:1061 suppresses
- L8:770 excites
- L1:102 excites
- L9:2249 suppresses
- L5:330 excites opening parenthesis after a noun
- L2:450 suppresses
- L8:354 excites opening parenthesis before a citation year
- L3:57 excites
- L0:1622 excites
- L4:327 weakly suppresses
- L1:195 weakly suppresses
- L3:1388 weakly excites
Feeds later layers
- L16:264 strongly suppresses opening parenthesis in citations and asides
- L11:50 strongly suppresses
- L16:869 strongly suppresses
- L12:1287 strongly suppresses
- L18:682 strongly suppresses
- L17:1509 strongly suppresses
- L12:2933 strongly suppresses
- L16:1967 strongly suppresses
- L20:335 suppresses
- L20:1451 suppresses
- L17:304 suppresses opening parenthesis after a claim
- L19:889 suppresses
- L21:187 suppresses
- L20:104 excites
- L12:2745 suppresses
- L15:1374 suppresses
- L11:2191 suppresses unclear: brackets and parentheses in text
- L11:907 suppresses
- L15:2486 suppresses bracket or parenthesis inside a quote
- L25:213 suppresses
- L18:1365 suppresses
- L19:453 suppresses
- L19:2339 suppresses
- L15:608 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
- LETE
- レーション
- rene
- roperty
- Penguin
- 增值
- 从此
- 可靠
- isi
- contin
Suppresses
- 硐
- *);⏎
- 箴
- *)⏎
- &)
- "")⏎⏎
- &);⏎
- ]="
- ComVisible
- *)"