comma after a conjunction starting a clause
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. 3 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly 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 responds to a comma placed right after a joining word like "and", "or" or "yet" when a new clause or action follows, as in "and, taking the money". Its negative side is weak and not clearly meaningful. Turning it up gave text that did not clearly relate to this pattern.
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.
-
, or Red River; then following the course of Rio Roxo westward, to the degree of longitude 100 west . . . , then, crossing the said Red River, and running thence, by a line due north
-
always the same according to different sources and not always exact. So, how do you know what information to believe? It is all very well reading books, but, have you ever considered that you will never get a sense of shape and size from
-
exchange it, what has been exchanged as well as its substitute is sacred. It is not to be redeemed."⏎Then let these things be exchanged for money, and, taking the money in your hand
-
is perpetrated onto someone as in the story of “St George and the Dragon.” However, we can redeem the evil dragon by turning him into a protector, but, again, he would have to protect someone or something.⏎From St George to George
-
human exploration of the solar system and beyond. In that expectation, we were cruelly disappointed. Partisan politics, a collective failure of will and ambition, and, yes, US victory in the Cold War all played a part, because the US
-
amusement.⏎There was a half-decayed log lying by the roadside, and Grant sat down on it, calmly pulled a cigar from his cigar case, and, requesting one of his staff who stood nearby to furnish him a sheet of paper,
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: a number of the same number of the same number of the same number of the same number of the same number of the same number of the same number of the same number of the same number of the same number of the same number of the
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
measured unless subject to both ethical and commercial consideration.⏎Home page url⏎Download or read it online for free here:⏎by Flora Klick
-
algae, such as common⏎species of Sargassum.⏎Hawai‘i: Northwest Hawaiian⏎Islands, O‘ahu, Maui, and Hawai‘i Island.⏎Worldwide: Eastern Atlantic,⏎Caribbean, Indian and Pacific O
-
the three descriptions.⏎A. Ask a volunteer to leave the room. When s/he returns, s/he will be asked to see if s/he can guess an<|endoftext|>
-
the raku firing process result in exciting colors, surfaces, and metallic luster effects.⏎|Presenter:||Lori Mills (Faculty)|⏎|Location:||Outside NE Corner Tower|⏎|Time:||9:30 am
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
- L2:409 suppresses
- L1:38 suppresses
- L2:564 excites
- L2:1283 suppresses
- L2:375 suppresses
- L0:2370 excites
- L2:1036 weakly excites
- L0:11 weakly excites
- L2:267 weakly excites
- L2:173 weakly excites
- L1:2648 weakly excites
- L0:899 weakly excites
- L0:1563 weakly excites
- L1:807 weakly suppresses
- L1:1 weakly excites
- L2:45 weakly excites
- L1:34 weakly excites
- L0:1626 weakly suppresses
- L1:349 weakly excites
- L1:8 weakly excites
- L2:1427 weakly suppresses
- L2:1534 weakly excites
- L1:3033 weakly excites
- L2:1322 weakly suppresses
Feeds later layers
- L9:268 excites
- L4:260 excites
- L4:155 suppresses
- L4:43 excites
- L6:660 excites
- L10:704 excites
- L7:876 excites
- L5:107 weakly suppresses
- L8:1468 weakly excites
- L5:44 weakly suppresses
- L4:5 weakly suppresses
- L6:258 weakly suppresses
- L4:559 weakly excites
- L6:491 weakly excites
- L9:802 weakly excites
- L15:933 weakly suppresses
- L6:2652 weakly excites
- L12:1196 weakly suppresses
- L13:1243 weakly excites
- L7:589 weakly excites
- L7:222 weakly suppresses
- L7:85 weakly excites
- L8:73 weakly suppresses
- L15:1298 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
- allis
- izzato
- vides
- Study
- urry
- olio
- Curso
- Studies
- ologies
- Read
Suppresses
- 当时
- ำ
- whim
- 凼
- ]) ⏎ ⏎
- ảo
- 车上
- 当时的
- 机器
- 市场的