'of' after a noun phrase, as in 'of course'
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. 16 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 the word 'of' when it follows a noun and sets up a phrase like 'of course', 'each of which', or 'one of each'. Turning it up or down did not produce a clear theme in the model's writing.
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.
-
at AATR to help us through cycling in the wind either on solo or small group rides.⏎Plan your route!⏎If you’re heading out for a ride of reasonable length, check the weather forecast and wind direction for the duration of the ride
-
has a population of 4,008,994. Popular nicknames for Philadelphia are Philly and The City of Brotherly Love, the latter of which comes
-
across disparate operating systems.⏎Kubernetes clusters are actually a set of nodes categorized into master nodes and worker nodes. Every cluster needs to have at least one of each. While the master node is meant to control the state of a cluster,
-
ating a gas balloon similar to the ones used by Lowe, and allowing visitors to interact with Lincoln, Lowe and Mary Henry themselves (portrayed by actors, of course). There will also be presentations about ballooning and espionage during the war,
-
of materials and weight and requires something that is much harder to come by: a propellant. You can make your own gunpowder, of course, but refining the process to create effective munitions is as tricky as building
-
9 determined that enslaved people in the state would be emancipated on July 4, 1827, but there were addenda and provisos, of course, so not everyone was instantly manumitted on the magic date, and
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: of of of, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and, and,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
why it is so significant that the angels described Jesus’ birth as “peace on earth.”⏎9 And an angel of the Lord appeared to them, and the glory of the Lord shone around them, and they were filled with great fear.
-
3,4 The voice of the LORD is upon the waters; The God of glory thunders, The LORD is over many waters. The voice of the LORD
-
content make, and they explicitly disclaim, any express or implied representations or warranties of any kind, including, without limitation, representations and warranties as to the functionality of the translation feature or the accuracy or completeness of the translations.⏎Translations are not retained
-
irling wheels⏎Four strange whirling wheels are associated with the cherubim.⏎Then I looked, and behold, in the expanse that was over the heads of the cherubim something like a sapphire stone, in appearance in the likeness
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
- L3:315 excites 'of' after a noun phrase
- L4:4 suppresses
- L1:725 suppresses "to" in "referred to"
- L6:357 suppresses
- L9:2264 excites 'of' after superlatives and 'the X of'
- L15:785 suppresses 'of' in set phrases like 'of course'
- L5:363 excites
- L4:1686 excites
- L0:40 suppresses 'of' in 'sort of' hedges
- L0:2726 excites
- L0:1220 suppresses
- L2:895 suppresses
- L0:117 suppresses
- L6:13 excites
- L2:422 weakly suppresses
- L3:1379 weakly excites
- L8:16 weakly suppresses
- L1:2567 weakly excites
- L2:1795 weakly suppresses
- L2:202 weakly excites "of" after "many" or "which"
- L6:703 weakly suppresses
- L0:897 weakly suppresses
- L7:122 weakly excites
- L1:316 weakly suppresses
Feeds later layers
- L17:317 strongly excites 'of' after a noun phrase
- L17:1835 strongly excites
- L18:2808 excites
- L19:715 excites
- L17:1686 excites
- L22:229 excites "of" after a noun in "X of Y"
- L17:1177 excites
- L20:283 excites
- L17:2297 excites
- L19:2893 suppresses
- L25:231 suppresses
- L18:293 excites
- L17:963 excites
- L19:2138 weakly excites "of" in "matter of" phrases
- L19:1992 weakly excites
- L17:1507 weakly excites
- L20:61 weakly suppresses
- L17:1246 weakly excites
- L20:64 weakly suppresses
- L21:1287 weakly suppresses
- L22:496 weakly excites
- L18:2244 weakly excites "of" inside titles and names
- L27:146 weakly excites
- L20:1972 weakly excites 'of' in 'of course'
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
- sorts
- screen
- 正しい
- screens
- contr
- urn
- fact
- _BOOL
- liche
- course
Suppresses
- стоя
- voke
- "]];⏎
- zsche
- ��
- ventory
- etest
- 谈起
- bies
- ons