'of' inside fixed noun phrases
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. 13 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" in set phrases like "field of view", "cause of death" or "conflicts of interest", where the next word is fairly predictable. Turning it down made the model get stuck repeating "of the year of the year", so it seems to help keep phrases like this flowing.
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.
-
time during a comets apparition even⏎binoculars will be too powerful...⏎A telescope is a big mistake, it will have much too narrow a field of view.⏎Please, you should not just go out to an unfamiliar rural "dark
-
em examination, please contact the Social Worker at the State Coroner's Office.⏎In most coronial cases a post-mortem is required to determine the precise cause of death with certainty. However, on some occasions the Coroner can determine the cause
-
way, Gao and his colleagues demonstrated that the tool could reveal signs of pulmonary fibrosis in COVID long-haulers, thus helping to account for shortness of breath, coughing and other lung troubles. A diagnosis, he suggests, that
-
5i2.5⏎© 2018 Author. This work is licensed under the Creative Commons Attribution 4.0 International License.⏎Conflicts of Interest: None⏎Mild traumatic brain injury is defined as a traumatically
-
the world? 70% of all PET bottles are recycled here, compared to 31% in the USA. But before you give yourself a pat on the back for this achievement, can you name the invisible force behind it? It
-
actually delivering on the goal of limiting warming to 2°C. The authors drew on available data to establish global reserves at 1,294 billion barrels of oil, 192 trillion cubic metres of gas, 728
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 my own of my of my of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
nests, but the Son of Man has no place to rest."⏎American Standard Version⏎And Jesus said unto him, The foxes have holes, and the birds of the heaven have nests; but the Son of man hath not where to lay his
-
atuation for someone, especially someone inappropriate or unattainable. … Unlike crushes and states of infatuation, love truly sees and accepts their object of affection. Love is an intense feeling of deep affection
-
about cartilage damage is a good place to start. There are times when cartilage damage is so significant that replacing the joint is necessary to help restore function and range of motion.⏎What Are the Primary Functions of Cartilage?⏎Cartilage plays a
-
Holman Christian Standard BibleJesus told him, “Foxes have dens, and birds of the sky have nests, but the Son of Man has no place to lay His
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
- L11:1392 strongly excites 'of' in fixed technical phrases
- L11:2057 strongly excites
- L12:2176 excites 'of' after a place or role noun
- L8:1673 suppresses
- L12:2046 suppresses
- L11:2424 suppresses
- L11:1158 suppresses
- L12:1423 suppresses
- L12:1579 suppresses
- L11:2840 excites 'in' after a size measurement
- L8:1801 suppresses
- L8:202 excites
- L9:2196 excites
- L12:1116 suppresses 'of' in formal noun phrases, mixed
- L12:1053 suppresses
- L11:1274 suppresses
- L12:1126 excites
- L12:1984 suppresses
- L10:2377 excites
- L8:1541 excites
- L10:564 excites
- L11:173 suppresses
- L4:990 suppresses
- L12:1518 suppresses
Feeds later layers
- L17:1771 strongly suppresses
- L17:546 strongly suppresses
- L17:1241 suppresses
- L14:1737 suppresses
- L16:1855 suppresses
- L14:1037 suppresses
- L16:2715 suppresses
- L14:1259 suppresses
- L14:1393 suppresses
- L19:2095 suppresses
- L23:2504 suppresses
- L16:2743 suppresses
- L15:435 excites
- L16:302 suppresses
- L16:961 suppresses
- L14:539 suppresses
- L17:2463 suppresses
- L17:2736 suppresses
- L18:2467 suppresses
- L17:375 excites
- L20:2599 suppresses
- L16:434 suppresses
- L16:2443 suppresses
- L22:2882 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
- 况
- ippers
- etypes
- ector
- ifiable
- quake
- emet
- 畿
- 平台上
- ifications
Suppresses
- -of
- _of
- Of
- Of
- _Of
- of
- _OF
- зрения
- interest
- concern