unclear: word fragments like 'de' in 'deities'
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 fairly consistent: several of them 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 small pieces of longer words, most often the piece "de" in "deities". The pattern is weak and mixed, and the text it produced when turned up or down showed no clear theme.
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.
-
a drop of the heart's blood of Mother Earth creates fertility according to the Orient region. French jewelers called ruby 'the gem of gems' or the ‘dearly loved stone’. Ruby is spoken of in the book of Job, in
-
represent the venue itself in a traditional and respectful way. A theatre building that connects with something as ancient as Chinese acrobatics itself was the outcome.⏎As was the peak of Chinese architecture, during
-
about as you read?⏎Mar 07, Minutes Buy. Anthony Lukas Award A landmark work of science, history and reporting on the past, present and imperiled future of the Great Lakes. But they are under threat as never before,
-
belief 1n winged minor gods but the concept of the intermediary, wholly good angel was most fully developed in the West. Where there existed a thousand little deities, some kind, some mischievous, but all approachable and bri
-
repertory.⏎The Symphony might be regarded as a battle, a study of opposites in conflict: the stately versus the agitated, the singable diatonic melody against the sinewy chromatic theme, the foreboding
-
h-only" party, which insisted that no other god but the Lord must be worshiped, and the more cosmopolitan leaders who favored incorporating other national and tribal deities into public worship. In the north, a Yahweh-only coup had
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: if the end-tempered and the end-tempered if the end-tempered if the end-tempered if the end-tempered if the end-tempered if the end-tempered if the end-tempered
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
New York factory, which was a one-piece brass measuring 33.5 meters (109’) tall. This flagpole is still the tallest in service at An
-
gent plate margin between the African Plate and the Eurasian Plate. It is the tallest active volcano in Europe, currently standing 3,329 m (10,922 ft) high, though this varies with summit eruptions
-
conflict between the shame and guilt associated with the act of rape. In conclusion, Williams observes that Lucrece may be best understood as Shakespeare's portrayal of the archetypal rape victim. Edward T.
-
cm⏎Possibly William Bryant, sale at Edward Foster’s, London, 15th May 1823 as ‘Holbein’⏎Archibald Alexander Leslie Melville, 13th Earl of Leven and
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
- L22:151 weakly suppresses
- L19:647 weakly suppresses
- L13:1492 weakly excites
- L20:134 weakly suppresses
- L21:774 weakly excites
- L4:1924 weakly excites
- L15:2578 weakly suppresses
- L22:366 weakly excites
- L9:1925 weakly suppresses
- L7:799 weakly excites
- L3:108 weakly suppresses
- L8:2650 weakly excites
- L0:1727 weakly suppresses
- L7:887 weakly excites
- L2:492 weakly suppresses
- L21:654 weakly suppresses
- L22:1859 weakly excites
- L18:305 weakly excites
- L12:948 weakly suppresses
- L17:1703 weakly excites weak, mixed: 'on' and 'the' after publishing verbs
- L11:2553 weakly suppresses
- L17:3003 weakly suppresses
- L14:2829 weakly excites
- L14:1083 weakly excites
Feeds later layers
- L24:1898 weakly suppresses
- L24:2370 weakly suppresses
- L24:2645 weakly suppresses
- L24:1875 weakly suppresses
- L27:2462 weakly suppresses
- L24:2045 weakly excites acronym introductions
- L25:3042 weakly excites
- L26:682 weakly excites
- L26:488 weakly suppresses
- L24:1531 weakly suppresses
- L24:2857 weakly excites
- L24:776 weakly suppresses
- L25:2502 weakly excites
- L26:2550 weakly excites
- L27:178 weakly excites
- L27:2469 weakly suppresses
- L25:726 weakly suppresses
- L25:1514 weakly suppresses
- L26:2289 weakly excites
- L26:841 weakly excites
- L25:1886 weakly suppresses
- L24:2263 weakly suppresses
- L24:1665 weakly excites
- L26:2782 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
- bidden
- 搏
- rous
- chner
- charges
- alar
- 新动能
- allo
- 限り
- 依法
Suppresses
- Ế
- эр
- #{@
- 份
- aturas
- (port
- 重伤
- itudes
- emet
- _PORT