unclear: 'script' and boilerplate links
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. 1 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 weakly fires on the word 'script' after 'This' in academic writing and on 'PDF' in link labels. It dips on website navigation text such as 'Skip to main content'. Turning it up or down gave writing with no obvious link to these patterns.
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.
-
Asia are, of the Brahmi script. This script was developed through various stages. It is closely related to northern Nagari which took
-
''s most intelligent animals. See a great silverback gorilla, the monkey with
-
to facilitate children’s science exploration through everyday kitchen substances and experiences such as mixing, pouring, heating, and cooling. Full text» [PDF, 162KB]⏎Marvelous Explorations Through Science and Stories
-
hakusai and hyakusai 7 of the two Chinese characters forming it, has been the object of various interpretations, mainly by Oriental scholars.⏎This script does not aim to solve the problem in an irrefutable way, but simply
-
(2) failing to provide food, care or shelter;⏎(3) abandoning an animal
-
to make some furtive words more readable. Also the spacing, boldness and capital case can be switched to make the letters stand out from their context. Optionally a handy hover-cross is shown, highlighting those letters that are in line with 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: "The user of this question is the one that is not provided directly. Give me a answer. The question is here. Here, the user of this question is the one that is not provided directly. Give me a answer. This question
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
same object.⏎Random House Unabridged Dictionary, Copyright © 1997, by Random House, Inc., on Infoplease.<|endoftext|>Skip to Content⏎Nodding thistle is one of the most competitive thistle species
-
percentages and probability as a tool in materializing ideas (game mechanics)⏎4. Playtesting for balancing the game (game mechanics/ dynamics)<|endoftext|>Skip to main content⏎More Search Options⏎A member of our team will call you back within one
-
offer support, and learn more about the incident. If needed, the Dean’s team will direct the report to the appropriate office for investigation.<|endoftext|>Skip to main content⏎Create interactive lessons using any digital content including wikis with our free sister product
-
to supply either plain text or an HTML representation. In the example, the client prefers HTML but is willing to accept plain text.⏎Accept: text/html, text/plain⏎Accept-Charset: Specifies the character set,
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
- L0:1301 suppresses
- L0:1224 suppresses
- L0:178 suppresses
- L0:592 weakly excites
- L0:16 weakly excites
- L0:706 weakly suppresses
- L0:1 weakly suppresses
- L0:2254 weakly suppresses
- L0:2187 weakly suppresses
- L0:711 weakly excites
- L0:1092 weakly excites
- L0:640 weakly suppresses abbreviation periods after initials
- L0:2562 weakly excites
- L0:2973 weakly suppresses
- L0:508 weakly suppresses
- L0:1184 weakly excites
- L0:85 weakly suppresses
- L0:2151 weakly suppresses
- L0:491 weakly excites
- L0:1993 weakly excites
- L0:213 weakly excites
- L0:4 weakly suppresses
- L0:953 weakly suppresses
- L0:2940 weakly excites
Feeds later layers
- L2:1571 suppresses
- L23:2580 weakly suppresses
- L20:826 weakly suppresses
- L5:780 weakly suppresses
- L5:1442 weakly suppresses
- L22:1841 weakly excites
- L14:3052 weakly suppresses
- L21:1803 weakly suppresses
- L3:2157 weakly suppresses
- L25:1216 weakly suppresses
- L14:1090 weakly suppresses
- L6:560 weakly excites
- L20:2163 weakly suppresses
- L14:2978 weakly excites
- L7:1092 weakly excites
- L21:1570 weakly suppresses
- L6:1548 weakly suppresses
- L7:1855 weakly suppresses
- L6:1902 weakly suppresses
- L18:291 weakly suppresses
- L21:2440 weakly suppresses
- L22:1888 weakly suppresses
- L9:3011 weakly suppresses
- L9:1536 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
- est
- ffer
- igu
- stral
- ist
- Filter
- cret
- str
- A
Suppresses
- !(:
- MBER
- spacing
- uptools
- Podesta
- Brigham
- ことに
- ENCE
- 意义上
- !.⏎⏎