the word matter in no matter phrases
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. 17 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
It fires on the word "matter" in phrases like "no matter how", "no matter what" and "no matter where". It is very consistent across the examples. Turning it up made the model produce more "or" and "whether or not" style phrasing, which is a related idea of alternatives, though the outputs are otherwise mixed.
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.
-
Sekonyer village, central Borneo, remained poor no matter how much they burned surrounding forest to expand their farmland.⏎“Tragically, no matter how many resources are exploited
-
more kinds of plants in your yard, you will attract more animals, and help to increase your neighbourhood’s biodiversity.<|endoftext|>Want to beat your Scrabble friends no matter where you are?WordSolver Android App!⏎3 Definitions of Rifts⏎The
-
want whenever you want without having to work for it. It is about being convinced that you deserve things you want and someone has got to give them to you no matter what.⏎Below are what I believe are some of the pitfalls of this entitlement mentality
-
flare of a torch. They would scoop the substance from the spring by sopping it up with blankets and wringing the liquid out by hand into pots. No matter how many times they repeated that process, the spring would replenish itself with more
-
best—which can be very true—but I have found that it’s the simple things in life that teach our children the most.⏎Since he could barely walk, no matter where we were outside, my son Kendall was immediately drawn to sticks of every different
-
there may be better vision models in nature than the human eye movement. The insect or bird worlds probably have superior versions of eye movement than human eye movement.⏎No matter which animal or human example you use, the Creator got it right the first time
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 have been, at least,, at least,, at least, but not sure, most, but, likely, maybe, at least, but, at least, maybe, at least, and, to at least, most,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
. Excitingly, these histones, when expressed in E. coli, bind the genome nearly<|endoftext|>Definition of Nupe in English:⏎noun (plural same or Nupes)⏎2 [mass noun] The Benue-Congo language of
-
cool and moist. This is especially beneficial during the hottest, dryest months. During drought-like conditions, moisture can evaporate. Mulch will help to prevent or lessen this effect.⏎Weeds grow at a much faster rate than your garden’s
-
data table above.⏎Have you ever built an app based on Excel?⏎Method 1: The easiest way (VLOOKUP & MATCH)⏎Probably the easiest – or at least the shortest – way of returning values in 2 search dimensions is the
-
University of Chicago, the University of California, Davis, and the University of California, Berkeley.⏎As the Biden administration pledges to move the nation toward electric vehicles, or EVs, and states like California work to ban the sale of new completely gas
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
- L14:3046 excites
- L15:2935 suppresses
- L4:108 excites "or" offering a choice in web text
- L15:3059 suppresses 'and' or 'or' joining the last list items
- L4:556 excites
- L5:1203 suppresses
- L0:1177 suppresses
- L12:422 weakly suppresses
- L10:712 weakly suppresses
- L1:1187 weakly excites
- L14:3025 weakly suppresses
- L16:2436 weakly suppresses
- L16:71 weakly excites
- L15:633 weakly suppresses
- L13:2339 weakly excites
- L7:1917 weakly suppresses
- L13:104 weakly suppresses
- L8:2002 weakly suppresses
- L16:157 weakly excites
- L4:653 weakly excites "that" after "such" or "such a way"
- L8:2189 weakly excites
- L1:491 weakly excites
- L15:2508 weakly excites
- L11:91 weakly excites
Feeds later layers
- L19:2293 excites
- L19:289 suppresses
- L18:2517 excites
- L20:1334 excites
- L25:1270 excites
- L18:2362 excites
- L23:223 excites
- L18:2197 excites
- L18:1138 excites
- L27:376 excites
- L25:41 suppresses
- L18:2896 excites
- L20:1670 weakly excites
- L25:576 weakly excites
- L20:1599 weakly excites
- L21:728 weakly excites
- L19:1352 weakly excites
- L26:492 weakly excites
- L22:2765 weakly excites
- L18:3036 weakly excites
- L19:2700 weakly excites
- L22:1414 weakly excites
- L27:656 weakly excites
- L24:1071 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
- 或是
- 或多
- or
- 多么
- |h
- 有多么
- 何种
- whether
- 或者其他
- |i
Suppresses
- 且
- 只有
- 只能
- ailand
- 也只有
- 除了
- 两个
- ştir
- 尽量
- 不能再