court case names and 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. 22 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 the dot or name in court case citations, like 'Jones v. City of Opelika' or 'Johnson v. California'. When it was turned up, the model drifted into talk about courts and laws, and turning it down made it repeat phrases about women and people.
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.
-
mirroring any modern-day gun regulation, even for weapons that the people of that time could not have imagined existing?⏎Justice Breyer, joined by Justices Sotomayor and Kagan, focuses his dissent on the patent ludicrousness
-
with the study of literature.<|endoftext|>The two Supreme Court cases, Jones v. City of Opelika 316 U.S. 584
-
land… In short, every Indian nation is free to adopt its own laws and be ruled by them, to paraphrase the United States Supreme Court. Williams v. Lee, 358 U.S. 217 (19
-
system in an effort to survive inside the system; some inmates who would not separate themselves by race outside of a correctional institution. While the case of Johnson v. California challenged the practice of segregating inmates by using race as a classification, the
-
)⏎The two Supreme Court cases, Jones v. City of Opelika 316 U.S. 584 (1942) and 319 U.S. 103 (1943
-
4] Terre de Kerguelen, Appellee par M. Cook, Isle de la Desolation... [and] Plan du
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: for a good, well-EW, and even for a good, well-EW, and even for a good, well-EW, and even for a good, well-EW, and even for a good, well-EW,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Mr. Darcy while triggered by anger showed a level of independence that left him shocked and stunned.⏎The freedom she exhibited in finally accepting him in direct defiance of Lady Catherine and knowing her father would disapprove was unusual even for Austen. In
-
Cachefly for the bandwidth for this show.<|endoftext|>The Scarlet Letter is an 1850 romantic work of fiction in a historical setting, written by Nathaniel Hawthorne, and is considered to be his magnum opus.
-
—attended a gathering with her future husband, Percy Bysshe Shelley, her stepsister Claire Clairmont, Lord Byron and John Polidori. To pass the time, the group read a
-
her novel “Wide Sargasso Sea” — a creatively daring, strongly feminist, and brazenly anti-colonial counter to Charlotte Brontë's “Jane Eyre” — is considered an integral part of the literary canon, but
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
- L19:2006 suppresses
- L10:1278 excites
- L20:1020 suppresses
- L18:560 suppresses
- L20:2289 suppresses
- L9:3036 weakly suppresses
- L6:1161 weakly excites
- L4:1252 weakly excites
- L19:1333 weakly suppresses
- L20:2215 weakly suppresses
- L20:1377 weakly excites
- L2:2597 weakly suppresses
- L9:1293 weakly excites
- L2:2279 weakly suppresses
- L8:2911 weakly suppresses
- L20:551 weakly excites
- L10:1293 weakly suppresses
- L13:1134 weakly excites
- L18:1389 weakly suppresses
- L20:1089 weakly suppresses
- L18:361 weakly suppresses
- L6:1848 weakly excites
- L2:3013 weakly excites
- L13:1974 weakly excites
Feeds later layers
- L24:985 weakly excites
- L24:2502 weakly excites
- L25:2049 weakly excites
- L23:2727 weakly suppresses
- L23:939 weakly suppresses
- L23:2230 weakly excites
- L26:1772 weakly suppresses
- L26:1211 weakly excites
- L24:2994 weakly excites
- L25:1760 weakly suppresses
- L25:2854 weakly suppresses
- L25:920 weakly excites
- L25:2062 weakly suppresses
- L24:2439 weakly excites
- L25:1372 weakly excites
- L23:617 weakly excites
- L23:1430 weakly suppresses
- L25:1268 weakly suppresses
- L23:1719 weakly suppresses
- L26:2623 weakly suppresses
- L24:667 weakly suppresses
- L26:2509 weakly excites
- L27:835 weakly suppresses
- L23:2546 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
- icina
- ega
- 铟
- /antlr
- aminer
- Studio
- ano
- studio
- issippi
- orta
Suppresses
- ftar
- ージ
- imens
- ême
- ,ev
- ]]></
- acle
- crunch
- conde
- hid