unclear: mixed technical tokens
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. 24 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 reacts to a scattered set of spots in technical and historical text, such as the start of a name or a symbol inside a formula. Turning it up gave unrelated text with 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.
-
the syntax in configuration files is case-sensitive.⏎The machine configuration file, Machine.config, contains settings that apply to an entire computer. This file is located in the %runtime
-
the expressions of V(t), I(t), the trigonometric formula sin(X)sin(Y)=1/2(cos(X-Y)-cos(X+Y)), and the fact that (Vmax×Imax)/2=
-
(household power frequency in the United States), calculate impedances for all components and determine all voltage drops using Ohm's Law (E=IZ ; I=E/Z ; Z=E/I). The polar magnitudes of the results should closely agree
-
a cable connected to the ground, described the view of the capital city from high above. The Smithsonian Institution’s first secretary Joseph Henry, an advisor to balloonist Thaddeus Lowe, and his daughter Mary Henry observed the balloon’s controlled flight
-
10 seconds of the French folksong "Au Clair De La Lune" on April 9, 1860.⏎Édouard-Léon Scott de Martinville recorded the voice by using a "phonautograph"
-
2017, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5707683/#:~:text=It%20is%20a%20potent%20ant
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 different ways the government can create and supply the government. The idea that the government also just like a government also needs to realize the different ways of creating and supplying the government also, and also realize the different ways of creating and supplying the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
latonism, and it provoked a strong rebuttal from Bessarion.Converted from Orthodoxy to Roman Catholicism about 1427, Trebizond conducted his own school at Venice (1428-1
-
preserve General Jackson and this happy republic." ⏎- Junius Brutus Booth, Jr. (1821–1883) was married to Agnes Booth<|endoftext|>3.9 of 34 Ratings⏎Please rate this review
-
um Judaeorum, i.e., "fortified Jewish settlement". Jewish sources also mention the Jewish community in Nitra; for example, in his book Or Zarua, written in the late thirteenth century, a leading halakhic
-
Born||8 June 1236⏎|Died||1301 (aged 64–65)⏎Violant was born in Zaragoza, the daughter of King James I of Aragon (121
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
- L21:982 excites
- L23:3020 suppresses
- L23:852 suppresses
- L1:1340 weakly suppresses
- L23:312 weakly excites
- L23:777 weakly suppresses
- L23:1394 weakly suppresses
- L22:1294 weakly excites
- L7:3018 weakly excites
- L16:1053 weakly excites
- L11:645 weakly excites
- L14:973 weakly suppresses
- L18:701 weakly excites
- L17:445 weakly suppresses
- L21:50 weakly suppresses comma after 'So' or intro phrase
- L16:812 weakly excites
- L22:973 weakly excites comma after a full date
- L17:24 weakly excites
- L2:2357 weakly excites
- L20:1437 weakly suppresses
- L21:1792 weakly suppresses
- L2:169 weakly suppresses
- L2:2992 weakly excites
- L20:2882 weakly suppresses
Feeds later layers
- L25:246 weakly suppresses
- L25:311 weakly suppresses
- L27:263 weakly suppresses
- L25:232 weakly suppresses
- L25:617 weakly suppresses
- L26:611 weakly suppresses
- L25:2686 weakly excites
- L27:627 weakly suppresses
- L26:261 weakly suppresses
- L25:2704 weakly excites
- L26:1214 weakly excites
- L25:146 weakly excites
- L27:513 weakly suppresses
- L25:112 weakly suppresses
- L25:382 weakly suppresses
- L25:484 weakly excites
- L27:2316 weakly suppresses
- L27:186 weakly suppresses
- L26:1159 weakly suppresses
- L25:1756 weakly suppresses
- L26:342 weakly suppresses
- L26:2485 weakly suppresses capital I starting a word fragment
- L27:22 weakly suppresses
- L25:2757 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
- oured
- -grade
- seite
- -ended
- 통
- specific
- ancers
- obil
- حف
- ishment
Suppresses
- fy
- PERT
- син
- ?>:</
- mai
- APT
- .Fatal
- '</
- zm
- DDL