semicolons and list-like citations
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. 23 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 responds to places where a semicolon or a list of items is likely to come next, and to some bits of catalogue or citation text. Turning it up made the model output a long string of semicolons.
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.
-
Show all Grade Level⏎Resource Type: Activities⏎Results 1 - 7 of 7 matches⏎From Grid to Home⏎Lisa Gardiner, National Center for Atmospheric Research; Marie Johnson, United States Military Academy; Jonathan Hoffman, University of
-
you fold⏎This problem focuses on Dienes' Logiblocs. What is the same and⏎what is different about these pairs of shapes? Can you describe the⏎shapes in the picture?⏎10 space travellers are waiting to board their spaces
-
ology and healthcare accessibility. Formulation of regional guidelines may help reduce AOM burden.⏎- acute otitis media⏎Statistics from Altmetric.com⏎If you wish to reuse any or all of this article please use the link below which will take you
-
God and is profitable for doctrine, for reproof, for correction, for instruction in righteousness: That the man of God may be perfect, thoroughly
-
usev crater<|endoftext|>It is really amazing how Titanic still captures our imagination 100 years after she sunk to the bottom of the Atlantic in 1912.⏎I put this down to the fact that Titanic is firstly an amazing tale
-
James McIntyre⏎Read this poem in other languages⏎This poem has not been translated into any other language yet.⏎Still I Rise⏎The Road Not Taken⏎If You Forget Me⏎Edgar Allan Poe⏎Stopping By Woods On A Snowy
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: there is a type of research that research at the research and research research research research research research research research research research research research research research research research research research research research research research research research research research research research research research research research research research research research
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Vespidae) are widely distributed around most of the globe and are diverse and common in many landscapes (Beggs et al., 2011). Paper wasps are likely to influence many other species in terrestrial ecosystems because they
-
.⏎from the GNU version of the Collaborative International Dictionary of English⏎- n. Use; practice; exercise.⏎- transitive v. To use; to exercise; to inure; to accustom by practice.⏎from The Century Dictionary and
-
, 286,028; Lakewood, 143,754; Fort Collins, 124,665; Westminster, 103,599; and Pueblo,
-
comment.<|endoftext|>100 Chemistry Questions and Answers 8211; Basic Chemistrynbsp;nbsp;Chemistry Questions 8211; Learn Basic Chemistry Part 1 (1-25)nbsp;1
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
- L20:373 strongly excites
- L21:351 suppresses
- L19:1763 excites
- L20:206 suppresses
- L16:1598 excites
- L17:2027 excites
- L18:1465 excites semicolon joining clauses
- L19:2352 excites
- L22:381 weakly suppresses
- L4:2749 weakly excites
- L20:2986 weakly excites
- L17:1134 weakly excites
- L19:1020 weakly excites
- L10:1982 weakly suppresses
- L21:881 weakly suppresses comma after 'if' clause in prose, braces in code
- L2:1164 weakly excites
- L22:2647 weakly excites
- L21:1222 weakly suppresses
- L14:2182 weakly excites
- L21:2735 weakly suppresses "it" after whether or when
- L14:1790 weakly excites
- L13:2085 weakly excites
- L22:2098 weakly suppresses
- L5:2200 weakly suppresses
Feeds later layers
- L24:378 strongly excites
- L27:347 excites
- L26:421 excites
- L25:2637 excites
- L24:264 excites
- L25:318 excites
- L25:832 excites zero before a decimal point
- L25:296 weakly excites
- L26:603 weakly excites
- L26:1329 weakly excites
- L24:2812 weakly suppresses
- L24:2381 weakly excites
- L25:238 weakly excites
- L24:3025 weakly suppresses
- L25:511 weakly suppresses
- L25:461 weakly suppresses
- L24:2194 weakly excites
- L24:531 weakly excites
- L27:1985 weakly suppresses
- L27:1408 weakly suppresses
- L25:2289 weakly suppresses
- L27:634 weakly excites
- L24:962 weakly suppresses
- L25:2743 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
- ;
- ;
- ;
- ;b
- .;
- ;⏎
- ;c
- ;",
- ;"
- ;a
Suppresses
- ughter
- Matcher
- عد
- matches
- 才发现
- ####⏎
- .matches
- matches
- 等待
- Matching