'ch' starting a word fragment
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. 27 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 letters 'ch' when they begin the second half of a longer word, such as chandeliers, chivalrous or chieftain. When it was turned up, the model kept writing the word 'story', and when turned down it kept writing 'say'.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
Candles and devices using this oil-burning technique were the main means used to light homes, employed in candelabras, wall sconces, and elegant chandeliers.⏎It wasn't until the Victorian era that oil lamps improved, thanks
-
377. Edward, who loved knightly pursuits such as war, jousting, tournaments and hunting, surrounded himself with warriors, magnates, and chivalrous knights. The Hundred Years' War broke out as a result of a
-
!⏎2. Stop buying disposable plastic products⏎Single-use disposable plastic bottles are a big issue – whilst some people do reuse them for a limited time, more get chucked in the bin and head straight for our landfill.⏎There are lots of products
-
, Dutch and French.⏎New England: King Phillip’s War is the first supporting idea. * The New England Confederation put to the test during war with Indian chieftain King Phillip (Metacom)—Wampanoag Chief son
-
or "vain pledge," or thought to mean "having profitable pledges." The O'Boyles were chieftains in Donegal, ruling west Ulster with the O'Donn
-
silver plate, including crosses, chalices, and other spoils of the Church might be as much more.
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: say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say say
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
you would like to include in your essay. Next, group related ideas together and make sure they are ordered effectively. Finally, label your ideas using main headings and sub
-
Testaments, by Robert Jamieson, A.R. Fausset and David Brown at sacred-texts.com⏎neh 3:1THE NAMES AND ORDER OF THEM THAT BUILDED THE WALL OF JERUSALEM. (
-
also include idioms & quotations that either use this word or its translations in them or use any of the related words in English or Urdu translations. These idioms or quotations can also be taken as
-
well, tense.⏎At the most fundamental level, there are three tenses: past tense, present tense and future tense. Each of these tenses has four sub-categories (or aspects) which better describe the tense used within the writing. These
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
- L20:240 suppresses
- L18:1433 excites
- L19:1427 suppresses
- L24:109 suppresses
- L20:664 excites
- L19:2956 suppresses
- L22:939 suppresses
- L21:338 suppresses
- L19:2480 suppresses
- L22:260 excites
- L20:807 excites
- L18:2391 excites
- L22:204 excites
- L24:488 suppresses
- L18:1556 suppresses
- L20:2997 suppresses
- L19:2382 suppresses
- L21:2537 suppresses
- L23:119 excites
- L23:218 suppresses
- L23:1112 suppresses
- L19:2522 excites
- L19:592 suppresses
- L22:2089 suppresses
Feeds later layers
None: this is the last layer.
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
- story
- 忘
- aths
- story
- 文章
- 迹
- _story
- vive
- 空间
- TemplateName
Suppresses
- leg
- ям
- ~~
- say
- lü
- íc
- >Contact
- fare
- undry
- aux