'to' or 'into' showing movement between two things
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. 10 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 responds to small words like 'to' that link a starting point and an end point, as in 'from one place to another' or 'from the sea to dry land'. Turning it up did not produce a clear theme in the written text.
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.
-
train than to stop a 10-car passenger train.⏎Without that information, can you actually
-
that take this arithmetic thinking into account or, more recently, developing approaches to elementary school mathematics that build foundations of
-
moving from one location to another, they are stored in warehouses. At any given time, there are numerous
-
medieval Hungarian dialect, they still live in relative isolation, harvesting vegetables and nuts, some of which they exchange for oil, rice and other necessities brought in by village-to-village Romanian salesmen.⏎The Csangos fear that their culture and language
-
economy. We [need to] make sure that we're harvesting at a level that is sustainable into the future.⏎There is talk of wood-based biofuel competing against corn-based fuel relatively soon
-
way to elucidate the molecular basis of the genetic susceptibility to alcoholism.(JAMA. 1992;268:1877-1882)<|endoftext|>The Imaginative Treatment of Geometry in Waldorf
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: I like to do more with my time, and I have to make a decision to commit to a work-life balance and career path. So, I see a work-life balance career. So, I see a work-life balance career. I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
wedding bands, the baguette cut shape is a step cut diamond that is similar to other square to rectangular shaped step cuts, such as the emerald cut.⏎The baguette cut
-
opinion interms of the distribution of power and the management of societies. Sociologists theorize about collective forces at work. This review will take a communication perspective to public
-
and Leadership - October 20, 2016⏎- Rogue Bras to Bogarts: April Fool’s Day in the Media - April 1, 2
-
iga) in English.⏎As defined by the Cleasby & Vigfusson Old Norse to English dictionary:⏎- bröstuliga (brǫstuliga)⏎- adv
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
- L5:929 suppresses
- L9:1820 excites
- L6:800 excites
- L7:1762 excites
- L9:723 excites
- L8:719 excites
- L4:1941 weakly suppresses
- L4:574 weakly suppresses
- L5:3009 weakly excites 'to' after a verb or noun
- L6:2686 weakly excites
- L4:328 weakly suppresses 'to' after 'so as' and infinitive cues
- L9:2267 weakly excites 'with' after exchange nouns
- L8:2063 weakly suppresses
- L9:1449 weakly suppresses
- L5:49 weakly suppresses
- L8:956 weakly suppresses
- L7:2509 weakly suppresses
- L8:666 weakly suppresses
- L9:1114 weakly suppresses
- L9:2055 weakly suppresses
- L3:3020 weakly excites
- L8:250 weakly suppresses
- L9:274 weakly excites
- L8:809 weakly excites
Feeds later layers
- L11:2292 suppresses 'than' in comparisons
- L14:2301 suppresses
- L14:1688 suppresses
- L11:1141 suppresses
- L13:2931 suppresses
- L11:881 suppresses "with" after combine, mix or react verbs
- L11:1443 suppresses
- L13:1473 suppresses
- L16:1469 suppresses
- L14:1721 suppresses
- L13:1896 suppresses
- L16:686 suppresses
- L12:2014 suppresses
- L13:2356 suppresses
- L13:1678 suppresses
- L13:1419 suppresses
- L15:2247 suppresses
- L16:1269 suppresses
- L14:2821 suppresses
- L14:2634 suppresses
- L12:2150 suppresses
- L14:1421 suppresses
- L14:2854 suppresses
- L13:2173 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
- cased
- erox
- ohn
- unn
- igned
- .visual
- 实例
- unto
- عزل
- erry
Suppresses
- irement
- 宙
- igos
- otle
- oldt
- igo
- holders
- downfall
- eroon
- expectations