street and urban place 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. 6 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 responds to words for streets, avenues, districts and neighbourhoods, such as 'Avenues', 'Street' and 'shopping district'. When it was turned up, the model's text drifted toward city streets, street vendors and neighbourhoods in two of the four prompts.
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.
-
Ten store, and the destruction and looting began to spread east and west on 125th Street, targeting white-owned businesses between Fifth and Eighth Avenues. Some stores posted
-
more for products purchased in a shopping district with trees. Those same shoppers also say they are willing to stay longer and rate
-
, 77 Jackson Street, Saint Paul, Minnesota (Razed)⏎|Edit with form|⏎|Address:||77 Jackson Street|⏎|Neighborhood/s:||Downtown, Saint Paul, Minnesota|⏎|
-
825 by free working class African-Americans in uptown Manhattan. The area from W82nd to 88th Streets between Seventh and Eighth Avenues (click here for a larger version of the map) was
-
area. In 1917, an administration building was built on the block east of the temple and in 1972, the twenty-eight story LDS Church Office Building, which was, for many years, the tallest building in
-
1935, an employee at the Kress Five and Ten store at 256 W. 125th Street (just across the street from the Apollo Theater) caught 16-year-old black Puerto Rican Lino
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 first time the moon has been able to be seen. I think that's something about the problem. I think it's difficult to say. I think the best to go through, I can't help with, but the moon's been
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
Vintage, 1996.⏎Silver, Larry. “Nature and Nature’s God: Landscape and Cosmos of Albrecht Altdorfer.” Art Bulletin 81.2 (June 1999): 19
-
1,000 to 22,000 people.⏎If You Go⏎Doors to the Flagler College Auditorium — across from City Hall on Granada Street — open at 6 p.m. for lobby displays,
-
decades later, Houck is sharing that knowledge with visitors to Atlanta, taking them to places where prominent African
-
that if you take the average intelligence for all men and all women<|endoftext|>A metal pole anchored down with cables juts out of the sand next to a public walkway at First Landing State Park. Four small rods stick out horizontally from the sides.⏎
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
- L5:787 strongly excites
- L5:1422 excites
- L3:2077 weakly suppresses
- L1:2739 weakly suppresses
- L4:182 weakly suppresses
- L5:434 weakly excites
- L5:66 weakly excites
- L3:2073 weakly suppresses
- L3:1984 weakly excites
- L2:1657 weakly excites
- L2:2106 weakly suppresses
- L5:102 weakly excites
- L5:94 weakly suppresses
- L5:28 weakly excites
- L2:1477 weakly excites
- L2:49 weakly suppresses
- L1:2811 weakly excites
- L5:1354 weakly suppresses
- L4:150 weakly suppresses
- L5:169 weakly excites
- L4:1462 weakly excites
- L5:1401 weakly excites
- L4:1158 weakly excites
- L2:2692 weakly excites
Feeds later layers
- L10:1563 excites
- L9:2988 excites
- L8:2481 suppresses
- L7:410 excites
- L20:2861 weakly excites
- L10:2943 weakly excites
- L10:2255 weakly excites
- L7:1465 weakly excites
- L10:2909 weakly excites
- L16:2543 weakly excites
- L7:3023 weakly excites
- L8:2076 weakly suppresses 'make' before decisions or predictions
- L9:2586 weakly suppresses
- L7:1890 weakly suppresses
- L8:2132 weakly suppresses
- L8:2632 weakly suppresses
- L26:2247 weakly excites
- L7:2321 weakly excites
- L10:521 weakly excites
- L10:769 weakly suppresses
- L7:1945 weakly suppresses
- L7:657 weakly excites
- L7:1203 weakly excites
- L23:555 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
- street
- Snackbar
- Grove
- floors
- 位
- InSection
- 楼
- 街区
- getView
- -comm
Suppresses
- olidays
- öh
- omens
- TestData
- othy
- akening
- jdbc
- persistent
- .tbl
- Turks