'mil' or 'bil' inside names
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. 25 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 short syllable 'Mil' or 'Bil' when it starts or sits inside a name, like Milosevic, Milner, Milford or Tbilisi. Turning it up did not produce any clear change in the text that followed.
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.
-
if Western history isn’t littered with cravens and rapscallions.⏎As John Gallagher notes in Romania After Ceausescu, leaders such as Milosevic and Ceausescu used nationalism in ways that “denied individual
-
celebrates its 58th anniversary this year (Nature, 2009). In 2015, Stephen Hawking and Russian billionaire Yuri Milner announced the start of a well-funded program, ‘Breakthrough Initiatives’,
-
script: Gender and Policing in Peace Operations Chapter 15 The Guatemalan Peace Accords: Critical Reflections Chapter 16 Les Femmes Aux Milles Bras: Building Peace in Rwanda Chapter
-
ized according to diseases and regions, scientifically studied and provided with possibilities for use in modern medicine.⏎The museum is a training-scientific base for students of Tbilisi State Medical University and other higher education institutions, an information and resource center for
-
boat is a monument to human tenacity — or desperation. So is Joal. Fifty years ago, it was a village of 6,546 millet farmers and five pirogues. Then a terrible drought wrung the
-
explored the country in the 19th century. He ended a wandering life (he had earlier fought under Garibaldi in Italy) when he settled at Milford Sound in
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: what is the number of elements on the set on the number on the number on the number on the number on the number on the number on the number on the number on the number on the number on the number on the number on the number
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
protect the wheat land from Bedouins’ raids coming from the marginal lands to the east. This location is on the green line, the eastern border of the Moab Plateau.
-
دانشکده معماری و شهرسازی، دانشگاه علم و صنعت ایران، تهران، ایران
-
.⏎Metsäpohjainen piha paransi päiväkotilasten immuunijärjestelmää jo kuukaudessa<|endoftext|>Research Looks at Alternative
-
دانشکده معماری و شهرسازی، دانشگاه علم و صنعت ایران، تهران، ایران.⏎عنوان مقاله [English]⏎In the process of architecture
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
- L22:2199 suppresses
- L24:1002 excites
- L20:253 excites
- L23:1334 weakly suppresses
- L21:121 weakly suppresses
- L22:1328 weakly suppresses
- L23:3029 weakly suppresses
- L1:711 weakly excites
- L3:301 weakly suppresses
- L4:796 weakly suppresses
- L3:822 weakly excites
- L10:1921 weakly suppresses
- L7:2948 weakly excites
- L11:2992 weakly suppresses
- L23:506 weakly suppresses
- L14:97 weakly suppresses
- L10:763 weakly suppresses
- L7:1586 weakly excites
- L9:1059 weakly suppresses
- L7:1967 weakly suppresses
- L20:2685 weakly excites
- L19:1944 weakly excites
- L22:870 weakly excites
- L17:376 weakly suppresses
Feeds later layers
- L27:1027 weakly suppresses
- L27:813 weakly excites
- L26:1427 weakly suppresses
- L26:1143 weakly excites
- L26:760 weakly suppresses
- L26:2594 weakly excites
- L27:578 weakly excites
- L26:2350 weakly suppresses
- L26:676 weakly excites
- L27:1241 weakly suppresses
- L27:984 weakly excites
- L27:2614 weakly suppresses
- L27:2382 weakly excites
- L26:481 weakly suppresses
- L27:2475 weakly suppresses
- L27:2101 weakly suppresses
- L26:713 weakly excites
- L26:1965 weakly suppresses
- L26:2371 weakly excites
- L27:299 weakly suppresses
- L27:2607 weakly excites
- L27:2585 weakly suppresses
- L27:70 weakly excites
- L27:729 weakly 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
- 给她
- 把她
- 她的
- 因为她
- 她
- 对她
- 让她
- herself
- 但她
Suppresses
- riers
- aviolet
- u
- uars
- chrom
- igans
- .u
- uced
- perial
- orian