"of" after a noun in trademark and title 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
Fires on the word "of" right after a noun, especially in set phrases like "trademarks of", "fear of" and "a final report of". Turning it up made the model produce looping, repeated text.
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.
-
Hunter and Nancy A. Moran, will publish their paper, "Bacteriophages Encode Factors Required for Protection in<|endoftext|>Social media can create 'fear of missing out'⏎CHARLESTON, W.Va. -- As the constant
-
more.⏎- Perfect for center activities, homework and filling downtime between lessons⏎- Build critical reading skills⏎- A great review for testing⏎- Answers on back of cards for easy self checking⏎Figurative Language Practice Cards provide an easy way
-
example, in the image shown below, the bottle’s cast shadow is clearly incongruous with the shape of the bottle (as is the shadow on this cover of Time Magazine). Such obvious errors in a shadow are easy to spot, but more
-
1995-2012 Healthwise, Incorporated. Healthwise, Healthwise for every health decision, and the Healthwise logo are trademarks of Healthwise, Incorporated.⏎Find out what women really need.⏎Most Popular Topics⏎
-
in language use outside the classroom; using oral histories to promote literacy skills; using the newspaper with ESL students; and preparing the student for employment. A final report of the project is appended to the guide. (MSE) (Adjunct ER
-
⏎5 letter words⏎4 letter words⏎3 letter words⏎2 letter words⏎Images for ZEATINSLoading...⏎SCRABBLE is the registered trademark of Hasbro and J.W. Spear & Sons Limited. Our scrabble word finder
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 user, the community, and the school of the the school of the the school of the the school of the the school of the the school of the the school of the community, the government of the state, and the local government of
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
⏎Teacher Evaluation to Enhance Professional Practice 2000 US$ 24.95 167 pages<|endoftext|>Gate is the national monument of India. Situated in the heart of New Delhi⏎, it was designed by
-
of Publication⏎Blanks, C. M. (1948). San Antonio the Inland Gate. Retrieved from https://digitalcommons.pvamu.edu/pvamu-theses/395<|endoftext|>The Top
-
Cell Cycle and Growth by Extracellular Cues, Fall 2012. (Massachusetts Institute of Technology: MIT OpenCourseWare), http://ocw.mit.edu (Accessed). License: Creative Commons BY-NC-SA<|endoftext|>Make
-
Ten Nieuport 80s went to Brazil in November 1918, where they were used by the Escola de Aviacao Militar at Campos dos Afonsos. Brazil also received a number of Nieuport
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
- L9:1627 excites
- L10:2362 suppresses
- L3:315 excites 'of' after a noun phrase
- L8:2257 suppresses
- L5:363 excites
- L9:2717 excites
- L7:790 excites
- L9:2672 excites
- L9:2070 suppresses
- L4:1686 excites
- L8:1003 suppresses "the" or "of" before a noun phrase
- L9:2264 excites 'of' after superlatives and 'the X of'
- L8:968 excites
- L9:654 suppresses
- L10:352 suppresses
- L9:1266 weakly suppresses
- L5:91 weakly excites
- L10:299 weakly suppresses
- L2:852 weakly suppresses
- L10:669 weakly excites
- L6:1637 weakly excites
- L9:1550 weakly excites 'to' after 'matters' or 'means' phrases
- L9:1428 weakly excites
- L5:953 weakly suppresses
Feeds later layers
- L15:2633 suppresses
- L17:2267 suppresses
- L16:3007 suppresses
- L16:1096 suppresses
- L15:632 suppresses
- L12:850 excites
- L15:942 excites
- L14:2840 suppresses
- L12:1904 suppresses
- L16:2612 suppresses
- L14:1728 suppresses "of" after a numbered section or title
- L17:865 suppresses
- L16:2852 suppresses 'in' after a noun or participle
- L13:287 suppresses
- L16:961 suppresses
- L12:958 suppresses
- L12:1653 suppresses
- L18:183 excites
- L12:2535 suppresses
- L16:564 suppresses
- L14:1707 suppresses
- L12:2849 suppresses '-based' after a place name
- L13:844 suppresses 'of' after a percentage or share
- L16:1441 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
- אה
- tin
- vé
- arnings
- 石家
- Revenue
- érica
- ään
- -R
- żenia
Suppresses
- OfString
- of
- idual
- iculty
- utm
- of
- COPYRIGHT
- OF
- 场上
- ofday