'of' in fixed technical 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. 11 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 fires on the word 'of' in set phrases from technical or medical writing, such as 'degrees of freedom', 'field of view', 'cause of death' and 'center of gravity'. Turning it up made the model repeat 'of' over and over, which fits the same theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
movements in three-dimensional (3D) space is a complicated task, a challenge to both biological and artificial systems. Increasing the number of degrees of freedom of the arm makes this task more complicated (Hollerbach,
-
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
-
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 doctor will bend your injured knee and check for clicks, snaps and "catches" within the joint. Your doctor also will evaluate your knee's range of motion and will maneuver your knee to see whether your meniscus is sensitive to
-
trim-The angle with the horizontal surface of the water assumed by the float or hull under given conditions. trimming moment-The moment about a reference point, usually the center of gravity, exerted by the seaplane hull or float when held at a
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: "I have a dog is a red Apple watch 1000 dollars 20000 yen 30000円 10000円 10000円 100
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
is about the formative years of the fallen king. Out of print, but sometimes available from Alibris.⏎The Countess and the King: A Novel of the Countess of Dorchester and King James II by Susan Holloway Scott.
-
lest you forget the things your eyes have seen and lest they stray from your heart all the days of your life. And you are to make them known to your children and to your children’s children – the day you stood before the L-ord
-
I will make your seed to multiply as the stars of heaven, and I will give to your seed all these countries. And in your seed shall all the nations of the earth be blessed because Abraham obeyed my voice, and kept my charge,
-
oil. But if you don’t want to read through the study, just watch this video from the Renewable Fuels Association. They say a picture is worth a thousand words and in this case I think they’re right.<|endoftext|>Traf-O-
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
- L8:2385 suppresses
- L8:202 excites
- L8:1673 suppresses
- L5:2462 weakly excites
- L4:1369 weakly excites
- L6:2723 weakly excites "way of" before life, looking, eating
- L6:13 weakly excites
- L9:2985 weakly excites
- L6:2950 weakly suppresses
- L8:7 weakly excites
- L5:2269 weakly excites
- L0:1556 weakly excites
- L4:1347 weakly suppresses
- L5:2988 weakly suppresses
- L7:888 weakly suppresses
- L8:6 weakly suppresses
- L6:181 weakly excites periods in abbreviations and the 's in names
- L7:1970 weakly suppresses
- L2:2964 weakly suppresses
- L6:1995 weakly suppresses comma after a 'that' or 'who' clause
- L6:1745 weakly excites
- L9:335 weakly excites 'of' after nouns like anniversary or onset
- L4:1614 weakly suppresses
- L2:1534 weakly excites
Feeds later layers
- L13:1335 strongly excites 'of' inside fixed noun phrases
- L17:1771 strongly excites
- L14:1737 excites
- L17:546 excites
- L16:2715 excites
- L12:576 excites 'of' in fixed phrases like table of contents
- L23:2504 excites
- L26:711 excites 'years' after an age number
- L15:646 excites
- L22:142 suppresses 'of' after a quantity or noun phrase
- L13:1659 excites
- L20:2599 excites
- L14:1259 excites
- L18:2467 excites
- L16:1855 excites
- L12:2176 excites 'of' after a place or role noun
- L16:961 excites
- L23:567 excites
- L19:2622 excites
- L14:1265 excites
- L17:1241 excites
- L12:887 excites
- L13:700 excites
- L17:870 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
- -of
- of
- Of
- Of
- _of
- _Of
- _OF
- →of
- OfWork
- OF
Suppresses
- icolor
- 这些问题
- öne
- 年底
- elow
- enes
- št
- #ifdef
- licht
- egade