symbols in headings and titles
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. 11 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 decorative symbols and punctuation that sit inside titles or headings on web pages, such as hearts, arrows, bars and plus signs. It is negative on full stops and hyphens inside citations and codes. Turning it up gave unrelated text, so the steering tells us little.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
ly Jokes⏎☆ Cut and Paste⏎*Please look at the video preview above for a clear picture of the resource.⏎♥Why You’ll LOVE it!♥⏎Here is what teacher-buyers had to say about this resource!⏎♥ “
-
Magic Tricks for Scouts⏎> > > Triple Prizes < < <⏎piece of paper⏎- Write 1089 on the paper
-
[e^]k`t[~e]r*[=e]n"), n. [Cf.
-
AMA. To make a password even more secure, or applicable for services that require other characters, add special characters around it, like # JBS10AMA#.⏎Here are some
-
viflora. They have been sunk into E. plukenetii and are recognized as subsp. pl<|endoftext|>| ||Minding Your Mind || |⏎When Striving for Perfection Is a Problem⏎Last Reviewed by Faculty of
-
this article we explore 4 methods of how to conduct 2D lookups in Excel.⏎+++NEW+++NEW+++NEW+++ 2D XLOOKUP +++NEW+++NEW+++NEW+++⏎In a hurry?⏎Most common for 2
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: the difference between the number of the first year of the month when the first time the spring and summer of the year when the spring and summer of the year when the first year of the year when the car and the plant of the plant the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
osenblatt, David M., "Music Listening as Therapy" (2015). Loma Linda University Electronic Theses, Dissertations & Projects. Paper 294.⏎Loma Linda University Electronic Theses and Dissert
-
ography, July 2, 2014. DOI: 10.1007/s13524-014-0308-7<|endoftext|>Designs and simulation results are given for two
-
the problem was solved.⏎Joan Marie Galat, Pelican ©2012, 978-1-4556-1637-4⏎In the 21st century, with GPS and
-
Discover the world of gorillas and other primates their remarkable intelligence and social behavior⏎66 pages; ISBN 9780756668082⏎, or download in⏎More Juvenile Nonfiction⏎IB
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
- L9:102 excites
- L8:100 excites
- L7:284 suppresses
- L7:675 excites
- L6:582 weakly excites
- L8:991 weakly suppresses
- L9:284 weakly excites
- L0:2942 weakly suppresses
- L10:1431 weakly suppresses
- L9:1506 weakly excites
- L8:558 weakly excites
- L0:3054 weakly suppresses
- L7:969 weakly suppresses
- L10:312 weakly excites
- L2:2724 weakly suppresses
- L8:584 weakly excites
- L5:2161 weakly suppresses
- L2:1947 weakly suppresses
- L6:232 weakly excites
- L7:39 weakly excites
- L1:1340 weakly suppresses
- L4:1544 weakly suppresses
- L6:1779 weakly excites
- L5:313 weakly excites
Feeds later layers
- L13:525 excites second half of an opposite pair
- L13:1346 excites
- L15:1674 excites
- L14:2521 excites
- L17:2171 excites
- L17:1898 excites
- L19:358 suppresses
- L17:2056 excites
- L18:2276 weakly excites
- L14:1201 weakly excites
- L13:2230 weakly excites
- L15:1499 weakly excites
- L25:1420 weakly suppresses
- L16:1216 weakly excites
- L20:2145 weakly suppresses
- L21:1843 weakly suppresses
- L12:2886 weakly excites
- L15:1037 weakly excites
- L17:2486 weakly excites
- L19:2136 weakly suppresses
- L18:2982 weakly excites
- L20:2149 weakly excites
- L12:2020 weakly excites
- L19:2665 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
- aurus
- оборот
- ODO
- :\/\/
- plutôt
- icate
- @endsection
- peu
- quare
- OU
Suppresses
- ItemImage
- nants
- ture
- (ierr
- ippers
- nown
- ใหญ
- fäh
- screens
- ierte