opening parenthesis after a name or title
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. 18 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, title or heading, such as 'Elizabeth Woodville (c. 1439-1492)'. It goes the other way at the end of a sentence just before a new heading. Turning it up made the text repeat phrases like 'the role of the', so the steering does not show a clear theme.
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.
-
BMC Musculoskeletal Disorders volume 4, Article number: 4 (2003)⏎The purpose of this study is to present an automated system
-
Pretty Balloons (counting/number sentences) (Resource Book Only) eBookKindergarten|Grade
-
2. Stephen Ruane, 13. James Grealish, 14. John Grealish, 15. James Grealish (Michl), 16. John Carr (Pat), 17.
-
Simon, who owned horses and property, is skeptical about the longevity of the new parks.⏎“How are they going to make a park in the<|endoftext|>Elizabeth Woodville (c. 1439-1492) was the daughter
-
= 2 x 1/4 or 1/4 is 1/2 of 1/2<|endoftext|>Messier 52 (also known as M 52 or NGC 7654)
-
condition that you copy this document without modifications and in its entirety, including this copyright notice.⏎Book Two, Chapter Twenty-Four, "Man and the Evolution" (Part 2 of 3)⏎If indeed man has evolved in Nature out of
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 a . I have a a . I have a a. I have a a. I have a a. I have a a. I have a a. I have a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
by beautiful natural surroundings, making it an ideal place for meditation and spiritual relaxation. In this article, we will tell you about the history and architecture of the temple.⏎Dhyunsar Mahadev Temple History⏎The history of the temple is
-
is the average of these conditions over extended periods. This distinction is key in understanding current weather patterns, including the cold wave, within the broader context of climate change.⏎El Niño and Jet Stream Patterns: Unpredictable Weather Catalysts⏎Recent
-
the way you think in order to become more positive. By following these tips, you can start living a more confident life – no matter what your current situation is.⏎Negative self-talk is harmful and can hold us back from achieving our goals.⏎Negative
-
can be worn as a nearly invisible patch on the skin. The small size frees the patient from concerns with holding the pump in place and concealing it under clothing.⏎The insulin Nanopump, developed by Debiotech and industrialized by ST
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
- L17:1568 strongly suppresses
- L16:190 strongly suppresses end-of-document boundary token
- L16:234 strongly excites
- L17:1176 suppresses
- L15:2070 excites
- L14:1132 suppresses
- L5:203 excites
- L17:2848 excites
- L17:1906 excites
- L1:1842 suppresses
- L17:1024 suppresses
- L6:2439 weakly excites
- L15:1325 weakly excites
- L16:313 weakly excites
- L17:1316 weakly suppresses document boundaries and line ends
- L16:1411 weakly suppresses
- L4:658 weakly excites
- L12:1310 weakly excites
- L16:2189 weakly excites
- L15:987 weakly excites
- L0:2540 weakly excites
- L1:816 weakly suppresses
- L3:70 weakly excites
- L12:2961 weakly excites
Feeds later layers
- L19:550 excites
- L20:1162 excites
- L19:1940 excites
- L22:597 excites end of document token
- L19:1201 excites
- L19:975 excites
- L19:801 suppresses
- L21:186 suppresses
- L22:820 excites
- L19:909 excites
- L20:1487 excites
- L19:624 excites
- L26:449 suppresses
- L21:466 excites
- L24:11 excites
- L19:1069 excites
- L19:2615 suppresses
- L19:558 suppresses
- L25:397 weakly excites
- L21:264 weakly suppresses
- L23:374 weakly excites
- L19:2591 weakly excites
- L21:301 weakly excites
- L19:1479 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
- ().__
- ST
- ocal
- 三方
- قة
- وث
- abeth
- dist
- miss
- 硬
Suppresses
- ckeditor
- 儆
- DisplayStyle
- și
- Guaranteed
- Ariel
- ioutil
- ism
- elly
- idian