the 'inter' word beginning
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. 26 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 "inter" part at the start of words like interstates, interplay, interwoven and interrelated, and sometimes on a stray number. Turned up, the writing fell into loops of the word "down" or "single", which is not clearly related.
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.
-
of its market value. To the unprofessional eye the object might look good, but the truth lies in the details: use of the correct reversible materials; intrusive interventional methods kept to a minimum; careful documentation of the research and the work process
-
years Hartford has lost 10,000 residents every decade, and getting through and around the City by automobile is often frustrating and complicated. Two elevated interstates slice through Hartford, and its explosively growing suburbs come at the expense of
-
of Chhayavadi school of Hindi poetry which gained popularity in late 19th and early 20th centuries.⏎Kamayani depicts the interplay of human emotions, thoughts, and actions by taking mythological metaphors.
-
pier arches, consisting of ornamental trees and flowering shrubs, lay the burial grounds. The choir featured a colourful roof of red and green leafy trees interwoven with Indian laburnum flowers. Behind the ‘choir fence’
-
the validity and worth of the evidence, bias and prejudice become difficult to remove. These arguments are amplified by two interrelated aspects of 21st-century life: firstly, the “post-tr
-
a complex and inter-connected microbial neighborhood that appears to be associated with asthma, and akin to what has
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 do you do when your host hosters host themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves themselves
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
name Kudara 1. This name in Japan refers to the Korean state of Paekche⏎百濟 2, which at the time of the Three States (Samguk 三國) was occupying the South-Western part
-
Western pubblications they are designated with: "dw." and "jw." - they are abbreviations of Polish words: "dwuwieżowy" = twin-turreted and "jednowieżowy"
-
garden bed by using a garden fork or tiller to loosen the soil to a depth of 12 to 15 inches, then mix in a 2- to 4-inch layer of compost and/or aged manure.⏎Planting
-
berg, who -- like them -- had hung around CDC Corporation near the University of Washington cadging open time on the mainframe. Wennberg, founder of the Triakis Corporation, was then an electrical engineering student at the University of Washington. In
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
- L19:1752 excites
- L20:311 suppresses
- L22:2314 excites
- L7:520 excites
- L7:2731 weakly excites
- L16:161 weakly suppresses
- L2:2600 weakly suppresses
- L4:326 weakly suppresses
- L21:708 weakly suppresses
- L20:2616 weakly suppresses
- L4:776 weakly excites
- L0:1010 weakly suppresses
- L19:1042 weakly excites
- L17:391 weakly excites
- L16:860 weakly excites
- L17:1818 weakly excites
- L19:723 weakly excites
- L17:1594 weakly excites
- L1:2935 weakly suppresses
- L16:600 weakly suppresses
- L2:127 weakly excites
- L13:3021 weakly suppresses
- L21:2561 weakly excites
- L3:2629 weakly excites
Feeds later layers
- L27:2751 weakly suppresses
- L27:2021 weakly excites
- L27:698 weakly suppresses
- L27:704 weakly suppresses
- L27:1320 weakly excites
- L27:757 weakly suppresses
- L27:2199 weakly suppresses
- L27:860 weakly suppresses
- L27:1684 weakly suppresses
- L27:1800 weakly suppresses
- L27:3024 weakly suppresses
- L27:2526 weakly suppresses
- L27:2141 weakly excites
- L27:1749 weakly suppresses hyphen joining word parts
- L27:1460 weakly suppresses
- L27:2178 weakly suppresses
- L27:1958 weakly excites
- L27:1554 weakly suppresses
- L27:2047 weakly excites
- L27:1515 weakly suppresses
- L27:930 weakly suppresses
- L27:2142 weakly excites
- L27:2901 weakly excites
- L27:2152 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
- toa
- icit
- aina
- 一夜
- we
- PasswordField
- itm
- ếp
- boxed
- WAY
Suppresses
- yourselves
- themselves
- 所在
- &_
- しい
- noun
- النقد
- ьте
- 中关
- urile