'the' after named as or chosen as
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 'the' when it comes right after a phrase like 'named', 'chosen as', 'became' or 'made it', where a title or superlative noun is about to follow. Turned up, the model's text drifted into unrelated topics and lists.
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.
-
25 featured films to be preserved in the Library of Congress’ National Film Registry, and in 2008, the American Film Institute named it the greatest animated film of all time.⏎Disney film historian J.B. Kaufman,
-
leading and supporting students as they decide on a major and eventual career path. She also knows how to match students with high quality internship opportunities. She has been named the Harry W. Rodgers, III Distinguished Teacher of the Year (Loyola
-
- About this particular statue of the Infant KingHow did this image come to be chosen as the symbol of, and the object of, devotion at the Shrine of Christ the King
-
within five years. Part V describes the importance of the Metales y Derivados case for policy and community activism across borders, as well as its status as the poster child for the failure of NAFTA as a model for protecting public health and the
-
productivity of H. cervicornis. H. musciformis’⏎successfully invaded into the same niche, and replaced H. cervicornis⏎as the species most often found attached or entangled on the upper branches<|endoftext|>This site is
-
its early days as a directory for college students. In addition to personal pages messaging, it handles groups, events, and a selling marketplace. Hollywood even made it the subject of the movie The Social Network. By the way, social networks are 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: " ' ' " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " "
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
spontaneous outburst of violence in Tulsa.⏎In “Wilmington’s Lie: The Murderous Coup of 1898 and The Rise of White Suprem
-
, and sorghum.⏎The names laetrile, Laetrile, vitamin B-17, and amygdalin are often used in place of one another, but they are not the same product. The chemical make-up of La
-
posed James II as England’s rightful king and cutting access to the lucrative Spanish trade.⏎England, Holland and Prussia supported the Habsburg Empire’s rival claimant to the Spanish throne, Archduke Charles. In 1701
-
economic system caused Albania to lag far behind its neighbors in terms of economic development.⏎After Hoxha died in 1985, his hand-picked successor, Ramiz Alia, who became party leader while retaining his post as titular
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
- L5:2448 suppresses
- L11:104 excites
- L11:2662 suppresses
- L11:1373 suppresses
- L5:535 excites
- L9:1249 suppresses
- L6:2604 suppresses
- L11:505 suppresses
- L8:1650 excites
- L10:1504 suppresses
- L6:2079 suppresses
- L11:679 excites
- L9:1553 weakly suppresses
- L10:1063 weakly suppresses
- L9:608 weakly excites
- L11:1450 weakly suppresses
- L10:2601 weakly excites
- L8:2764 weakly suppresses
- L11:2778 weakly suppresses
- L11:2994 weakly excites
- L11:1793 weakly excites
- L11:1649 weakly excites
- L7:1229 weakly suppresses
- L10:673 weakly suppresses
Feeds later layers
- L15:2387 suppresses
- L14:2001 excites
- L14:321 excites
- L14:1285 suppresses
- L17:3031 excites
- L14:2140 weakly excites
- L16:2111 weakly excites
- L20:1696 weakly suppresses
- L14:2885 weakly suppresses
- L16:821 weakly excites
- L18:2995 weakly excites
- L17:240 weakly excites
- L17:636 weakly excites
- L16:1771 weakly excites
- L19:1967 weakly excites
- L21:1008 weakly excites
- L21:1211 weakly suppresses
- L13:1655 weakly excites
- L18:2347 weakly excites
- L14:3012 weakly excites
- L16:1492 weakly excites
- L24:2914 weakly suppresses
- L25:1468 weakly excites
- L22:2930 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
- 替代
- ael
- opic
- eer
- orph
- 更换
- asting
- 嬗
- 取代
- TextEditingController
Suppresses
- nữa
- ]+\
- inc
- 还包括
- ince
- INCLUDED
- clud
- addItem
- Normal
- �