'of' after part-whole nouns
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. 12 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 phrases like "a subsidiary of", "a division of" or "a subset of", where a bigger group is about to be named. Turning it up did not give a clearly related change in the 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.
-
Mariana Mazzucato. Copyright © 2018. Available from PublicAffairs, an imprint of Perseus Books, LLC, a subsidiary of Hachette Book Group, Inc.⏎The January 2010
-
913, upon passage of the 16th Amendment (which created the federal income tax), total government spending at all levels was about 8% of gross domestic product. By 1940, it was approximately 1
-
Highway Administration, a division of the U.S. Department of Transportation, has released the newest numbers for miles driven
-
the inventory of words in a particular language.⏎It is now widely accepted that lexicography is a scholarly discipline in its own right and not a sub-branch of applied linguistics, as the chief object of study in lexicography is the
-
or Computer Vision(how Google Lens works).⏎Machine Learning is the field of study that will enable computers to learn without being explicitly programmed.⏎Machine Learning is a subset of Artificial Intelligence.⏎Common ML Techniques⏎There are broadly three types of machine learning;
-
the ocean are threatening to the ecosystem, but all is not gloomy. This is where BLUE BEAUTY comes to the rescue. It’s a sub-section of clean or green beauty but dominantly focuses on the well-being of the<|endoftext|>Memory
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: the new and the old. The new and the old. The new and the old. The new and the old. The list. The new. The new. The new. The new. The new. The new. The new.
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
spot and demonstrate limited growth are generally considered to be benign.⏎More dangerous, or malignant, tumors form when two things occur:⏎a cancerous cell manages to move throughout the body using the blood or lymph systems, destroying healthy tissue in a process called
-
-resistant container. Keep Ampicillin out of the reach of children and away from pets.⏎Do not use Ampicillin if
-
ecost, the Spirit of the Promise was poured out on the disciples, gathered "together in one place."92 While awaiting the
-
respiratory diseases. The aerosol also contains particles that can cause inflammation of the lungs.⏎The use of nicotine in vaping products can cause physical dependence. Nicotine crosses the bloodstream and activates the
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
- L11:437 suppresses
- L6:2217 suppresses
- L11:2547 suppresses
- L9:2566 excites
- L10:2362 suppresses
- L10:132 excites
- L6:2842 excites
- L11:1858 suppresses
- L7:1152 suppresses
- L6:800 excites
- L9:2545 suppresses
- L10:1429 suppresses
- L11:506 suppresses
- L10:2468 suppresses
- L10:1649 excites
- L6:1004 excites
- L10:2138 weakly suppresses
- L8:1428 weakly excites 'on' introducing a location or surface
- L5:1477 weakly excites
- L11:2860 weakly suppresses
- L9:2390 weakly excites
- L11:2207 weakly suppresses "of" after a noun in trademark and title phrases
- L9:1858 weakly excites
- L10:1603 weakly excites
Feeds later layers
- L15:1338 excites
- L15:1514 excites
- L16:1984 excites
- L14:1707 excites
- L20:1038 excites
- L16:1798 excites
- L16:3007 excites
- L16:1096 excites
- L14:1508 excites
- L16:1051 excites
- L17:2959 excites
- L16:2789 excites
- L13:2135 excites
- L15:1741 excites
- L15:1305 excites
- L17:2022 excites
- L17:189 excites
- L18:2573 excites
- L22:1993 excites
- L17:2620 excites
- L17:1287 weakly excites
- L16:1128 weakly excites
- L15:1127 weakly excites
- L17:1888 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
- 总
- cum
- itle
- urs
- Amazon
- uded
- 總
- ified
- I
- ful
Suppresses
- lashes
- qx
- 砣
- grave
- 让她
- zac
- 痘
- quine
- qx
- 套房