'of' after a noun phrase head
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. 7 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 lights up on the word 'of' when it follows a noun like 'set', 'type', 'part', or 'style', as in 'a set of criteria'. Turning it up made the model's text drift into unrelated topics like code and weather, so the steering does not show a clear effect.
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.
-
full information.)⏎Welcome to English 303: Development of technical writing skills and styles. This course requires a different set of criteria from that used in composition courses, both in teaching and in evaluating students’
-
bones and the tailbone (coccyx).⏎- Because this part of the spine bears the weight and stress of the trunk and upper body, the lum
-
used type of screw conveyor. Used to convey bulk materials from one part of a process to another
-
of Giving to Others⏎Generous Behavior Increases Happiness⏎Not only does giving to others increase their happiness, it also increases our own. Studies amongst different cultures and ages have consistently found that individuals who spend money on others report higher levels of happiness than
-
different set of binary codes than Fish and glass of red wine this is how to order food and
-
people who have good fitness levels. However, it is best to begin with Ashtanga yoga so you get a sense of how it feels to do that style of yoga. Yoga meditation is an integral part of the daily yoga practice. Yoga meditation
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: what is the name of the place where the two are, the thing that is the universe, and the universe is everything. Let me think of the answer, the place where we are. So, the answer to this is everything. So
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
American institution, both feared and revered. Sadly, this savage tool that was once used to break the back of my ancestors finds its marks on too many
-
, which is vital for exchanging best practices, supporting afforestation events through volunteers from both sides and raising environmental awareness, stressing the urgency of collaboration on this global matter that affects everyone on the island.⏎The CV/AI methods
-
are 'lime mix' and 'canal' respectively. Hence it has been suggested that the name may mean a canal built of lime mix.⏎This Karaikal area formed an integral part of the Pallava kingdom in the eighth century(C 7
-
1619 and set an example for others to follow. As Virginians migrated in search of new opportunity, they took these lessons of self-rule they had learned and put them into practice in other colonies, territories and states.⏎In
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
- L2:725 suppresses
- L4:270 suppresses
- L3:194 suppresses
- L6:1556 suppresses
- L2:426 suppresses
- L5:2885 suppresses
- L5:1350 suppresses
- L2:2116 suppresses
- L3:1180 excites
- L5:2900 suppresses
- L3:472 weakly excites
- L1:1874 weakly excites
- L0:1344 weakly suppresses
- L6:2561 weakly excites
- L2:579 weakly suppresses
- L5:2566 weakly suppresses
- L3:1637 weakly suppresses
- L4:464 weakly suppresses
- L2:2278 weakly suppresses
- L6:1467 weakly excites
- L1:2925 weakly excites
- L6:345 weakly excites
- L4:633 weakly excites
- L5:1661 weakly excites
Feeds later layers
- L9:633 excites
- L11:1218 excites
- L18:1298 excites
- L12:1127 excites
- L8:272 excites
- L15:1187 excites
- L18:1425 excites
- L12:2722 excites
- L10:2418 excites
- L10:739 suppresses
- L8:706 excites
- L13:1004 excites
- L9:2829 excites
- L14:2444 excites
- L9:516 suppresses
- L13:685 excites
- L10:1115 excites
- L10:2403 excites
- L11:199 excites
- L14:1060 weakly excites
- L8:1137 weakly excites
- L13:2554 weakly excites
- L12:521 weakly excites
- L17:2139 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
- arity
- alent
- eel
- swana
- hape
- gain
- antage
- anse
- uality
- ategy
Suppresses
- ourselves
- przec
- 让我们
- 因为我们
- 跟她
- 给我们
- OnError
- 告诉我们
- กระบ
- -pack