'San' in 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. 24 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 lights up on the word 'San' when it begins a place name such as San Francisco, San Diego or San Marino. When it was turned up, the model's writing drifted into rambling, repeated phrases with no clear link to the neuron's topic.
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.
-
the meaty interior of a paste tomato. Some varieties can grow quite long like a horned pepper. Fleshy ones, like the D.O.P protected San Marzano
-
must develop culturally specific interventions. "The Chinese mass media, especially the mainstream media, is awash with<|endoftext|>Frida Kahlo and Diego Rivera Move to San Francisco⏎On November 10th, Frida and Diego arrive in San
-
2014 View and download them here.⏎Andrea McEvoy Spero‘s Ph.D. dissertation for the University of San Francisco, “THIS IS A PUBLIC RECORD”: TEACHING HUMAN RIGHTS THROUGH
-
) because the pumps are helping each other<|endoftext|>Canine fear-related aggression (Proceedings)⏎Nov 01, 2009⏎CVC IN SAN DIEGO PROCEEDINGS⏎Fear aggression is triggered by a stimulus that appears threatening
-
maps reflect expanding splotches of British-red.⏎One inquiring reader wants to know who controls the world's non-independent islands.⏎Detra K., from Los Gatos, California, wonders about the
-
Media Arts Center San Diego⏎With help from pharmaceutical students at UC San Diego, the Asian Pacific Health Foundation is continuing its campaign for a “Hepatitis Free San Diego.”⏎APH
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: an event at the heart of an intan (intan in full, an intan is an an integer an integer an integer) an intan is an intan an intan an intan an intan an intan an int
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
pressures associated with vibratory as well as hydrodynamic cavitation were measured. A stationary specimen subjected to vibratory cavitation and a hydrofoil of NACA 001
-
280nm and is therefore invisible to the naked eye. Naturally occurring UV-C rays from the sun are prevented from reaching the earth’s surface by the Ozone layer. However, they can be generated
-
-Paleogene extinction event killed off three-quarters of all species. It's widely believed now to have been caused by something slamming into the Earth - perhaps from the Oort Cloud - but there are competing theories that suggest a more terrestrial cause, or
-
. As the younger Hawaiian Islands push the Pacific tectonic plate downward, nearby Oahu<|endoftext|>In the era of 3D-printed guns, ammunition may prove
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
- L1:2668 weakly excites
- L1:631 weakly excites
- L21:2559 weakly suppresses
- L20:2435 weakly excites
- L2:1344 weakly suppresses
- L22:384 weakly excites
- L14:2923 weakly excites
- L15:765 weakly suppresses
- L18:664 weakly suppresses
- L18:2712 weakly excites
- L1:2947 weakly excites
- L17:429 weakly excites
- L18:1674 weakly excites
- L19:1032 weakly excites
- L14:5 weakly suppresses
- L23:1489 weakly excites
- L0:2908 weakly excites
- L19:2371 weakly excites
- L7:362 weakly excites
- L0:2760 weakly excites
- L6:1142 weakly suppresses
- L6:1736 weakly excites
- L5:707 weakly suppresses
- L16:2898 weakly suppresses
Feeds later layers
- L27:864 weakly excites
- L27:1257 weakly excites
- L26:2698 weakly suppresses
- L25:1358 weakly suppresses
- L25:1118 weakly suppresses
- L27:1870 weakly suppresses
- L27:2804 weakly suppresses
- L27:2165 weakly suppresses
- L27:1789 weakly suppresses
- L26:367 weakly excites
- L27:2675 weakly suppresses
- L26:1102 weakly suppresses fibromyalgia and hour counts
- L25:702 weakly suppresses
- L25:595 weakly excites
- L27:2234 weakly suppresses
- L26:1534 weakly excites
- L26:913 weakly suppresses
- L25:2736 weakly excites
- L25:1836 weakly excites
- L27:1427 weakly suppresses
- L25:821 weakly suppresses
- L26:559 weakly suppresses
- L25:1365 weakly excites
- L27:842 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
- eryl
- umba
- -group
- Group
- groupName
- Groups
- -groups
- AAC
- onor
- df
Suppresses
- aneously
- 整
- int
- 全
- artifact
- ละคร
- Neville
- Autom
- 艺术品
- 满