the piece 'Mont' 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 fires on the start 'Mont' of place names such as Montenegro, Montpellier and Montevideo. Turning it up made the model repeat 'a little a little', which has no clear link to the place names.
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.
-
Egypt, El Salvador, Finland, Grenada, Guatemala, Japan, Lao, PDR, Macedonia, former Yugoslav Republic of, Mongolia, Montenegro, New Zealand, Nicaragua, Nigeria, Norway, Pakistan, Paraguay,
-
16. A year later, the French settled in the area just north of present Montogmery and built Fort Toulouse on the river.⏎The Coushatta
-
istinguished Service Cross, the Croix de Guerre with palms, the French Legion of Honor, the Croce di Guerra of Italy, the War Medal of Montenegro, and ultimately, the Congressional Medal of Honor in March of 19
-
of yellowfin tuna (Thunnus albacares) in nearshore areas in French Polynesia, determined using ultrasonic telemetry⏎Centre IRD de Montpellier, BP 5045, 34032 Mont
-
Republic’s School of Agronomy through an agreement with the government of the province of Montevideo and the Primary Education Board.⏎The programme provides the opportunity for direct contact
-
was born in Montpelier, Vermont, on December 26, 1837
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 right and the left in the ricki ricki ricki ricki ricki ricki ricki ricki ricki ricki ricki ricki ricki rick
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
natural enemies that help suppress pest densities."⏎The study, "Multi-State Trials of Bt Sweet Corn Varieties for Control of the Corn Earworm (Lepidoptera: Noctuidae)," analyzed the performance of Bt sweet corn
-
of events. Fate defines events as ordered or "inevitable" and unavoidable. Classical and European mythology features three goddesses dispensing fate, known as Moirai in Greek mythology, as Parcae in Roman mythology, and as
-
prominent Creole family, the Puente de Ayala was opened to traffic in 1880. It connected the San Miguel district to Malate and Ermita.⏎The bridge also served Isla de Convalescencia, the
-
leshana tova (happy New Year), to send greeting cards, and perhaps plant trees in Israel in their honor.⏎Although the high holidays lack the merriment of Purim, the gift giving of Hanukkah, and the joy
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
- L20:435 excites
- L21:1720 excites
- L20:2520 weakly excites
- L18:1779 weakly suppresses
- L21:1379 weakly excites
- L19:1057 weakly excites
- L21:1555 weakly suppresses
- L18:1056 weakly suppresses
- L19:788 weakly suppresses
- L6:482 weakly suppresses
- L2:2055 weakly suppresses
- L0:655 weakly suppresses
- L7:3057 weakly excites
- L15:878 weakly suppresses
- L18:2042 weakly excites
- L1:1251 weakly suppresses
- L18:406 weakly excites
- L19:2972 weakly suppresses
- L18:1731 weakly excites
- L1:152 weakly excites
- L6:2547 weakly excites
- L16:2855 weakly suppresses
- L15:1308 weakly suppresses
- L1:1882 weakly suppresses
Feeds later layers
- L25:1586 excites
- L27:663 excites
- L27:797 excites
- L27:231 suppresses
- L25:45 suppresses
- L26:484 excites
- L26:411 suppresses
- L27:709 excites
- L25:1132 weakly suppresses
- L25:2182 weakly excites
- L26:213 weakly suppresses
- L27:854 weakly excites
- L26:1102 weakly suppresses fibromyalgia and hour counts
- L26:918 weakly excites
- L26:911 weakly suppresses
- L27:343 weakly suppresses
- L27:383 weakly suppresses
- L26:2923 weakly excites
- L27:2093 weakly excites
- L25:1048 weakly excites
- L25:2748 weakly suppresses
- L27:173 weakly suppresses
- L26:2293 weakly excites
- L27:1637 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
- a
- 一个
- asmine
- 一个小
- 一个问题
- ือน
- .makeText
- FFECT
- uality
- 是一個
Suppresses
- erto
- ientos
- BTS
- elier
- Gb
- URNS
- concentr
- Dana
- nämlich
- taire