opening parenthesis after a name
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 fires on the open bracket that comes right after a name or term, as in a full name followed by its short form or dates, like 'Point Calimere Wildlife Sanctuary (PCWS)'. Turning it up made the model write lists of numbers or bracketed options, which is only loosely related.
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.
-
<|endoftext|>Jacques-Louis David (1748-1825) is generally considered the leading figure
-
access to information and communication technologies (ICTs) can mitigate many issues in SIDS and help to diversify their
-
was aimed to investigate the patterns of liana-tree interactions in tropical dry evergreen forest of Point Calimere Wildlife Sanctuary (PCWS), south- east India. This study also analyzes the impact of various
-
temporoparietal junction (RTPJ), an area involved in mental state reasoning, is necessary for making
-
Silla (57 BC[note 1] – 935 AD) (
-
system for diagnosis of HIV infection (Weiss et al., 1985). As a result of this
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: by and as, and any and in the but and and and so and but and so and as and and and but and any and and and and and the and and and so and and and with and and and so and and and in
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
how these relate to what we know today.⏎In association with STFC Rutherford Appleton Laboratory⏎If your child has additional needs, please get in touch on [email protected] to discuss how we can best support you.⏎For information on our
-
anything with sugar in it⏎- anything with wheat, corn in it⏎- dog food⏎- cat food (mots of it)⏎What your Ferret<|endoftext|>The 2018
-
essay, we will delve into one of Mary Oliver’s poems, analyzing its themes, literary devices, and the impact it has on readers.⏎The Poem: [Insert Poem Title]⏎[Insert the full text of the chosen Mary Oliver poem
-
other fluids.⏎- Avoid exercising for at least 8 hours.⏎- 24 hours without alcohol.⏎- Have not smoked for 4 hours before the measurement (including e-cigarette with nicotine).⏎The measurement of basic metabolism in the diet office
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
- L18:980 suppresses opening parenthesis after a noun phrase
- L17:304 excites opening parenthesis after a claim
- L16:1513 excites opening parenthesis before a numbered list
- L17:238 excites
- L16:264 suppresses opening parenthesis in citations and asides
- L10:66 suppresses opening parenthesis after a citation or term
- L16:869 excites
- L16:710 suppresses
- L11:50 excites
- L17:1708 excites
- L18:1273 suppresses
- L18:682 weakly excites
- L10:1041 weakly suppresses opening parenthesis before phone or code
- L1:2526 weakly excites
- L14:90 weakly suppresses
- L5:330 weakly excites opening parenthesis after a noun
- L9:339 weakly excites
- L4:2309 weakly excites
- L16:183 weakly suppresses
- L17:2160 weakly suppresses
- L16:170 weakly suppresses
- L8:464 weakly excites
- L4:1799 weakly excites
- L17:35 weakly suppresses
Feeds later layers
- L20:36 suppresses
- L24:148 suppresses
- L24:35 excites
- L21:200 weakly excites
- L27:193 weakly suppresses
- L26:211 weakly excites
- L23:528 weakly excites
- L24:328 weakly suppresses
- L20:813 weakly suppresses
- L27:173 weakly excites
- L24:1690 weakly excites
- L20:1629 weakly suppresses
- L20:66 weakly suppresses
- L20:348 weakly suppresses
- L25:1270 weakly excites
- L22:267 weakly excites
- L21:478 weakly suppresses
- L25:476 weakly suppresses comma before 'but' in contrasts
- L27:19 weakly excites
- L22:160 weakly excites
- L27:78 weakly excites
- L27:289 weakly suppresses
- L27:139 weakly suppresses
- L21:524 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
- /
- .
- (
- /
- (
- ·
- ·
- q
- done
Suppresses
- 但由于
- ignant
- theless
- 但这
- uste
- }?
- 的同时
- ,**
- 但他
- mientras