"most" meaning "very" or "the most"
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. 19 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 word "most" when it means "the most" or "very," as in "at the most basic level" or "made it most useful." On the other side it fires on "so" in phrases like "changed so much." Turning it down made the model repeat "a lot of," which fits a theme of amounts and degrees.
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.
-
Now how does this hypothesis compare to that of Homer? Having lived almost four-centuries prior to Plato and nearly ten-centuries prior to Plotinus, Homer was most likely completely unacquainted with the ‘mind-body dualism’ theory held
-
and this was scarcely developed before the year 1746. The consummate hardness of the diamond, in spite of its high price, has made it most useful for purposes of grinding, polishing and drilling. Numerous attempts have been made to
-
a way of defining and solving problems by combining logic with insight, and by finding patterns and structure. At the most basic level, we use the abstract concept of “number” to understand what we
-
and Their Effectiveness⏎The recent approval of the commercial growing of Golden Rice (GR) based on guidelines set by a 5-interdepartmental circular is a most welcome, most awaited development.⏎Vitamin A deficiency (VAD) continues
-
too dim; however, it counts as sixth-closest to the Sun at 8.3 light years.⏎The seventh-closest is a star most denizens of Earth would recognise: Sirius (alpha Canis Majoris), with
-
, maintaining verb tense cohesiveness is important. However, understanding the different types of tense can make you, well, tense.⏎At the most fundamental level, there are three tenses: past tense, present tense and future
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: a lot of a lot a lot a lot a lot a lot a lot a lot a lot a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
contrast, an `io soaring over the Panaewa farm lots bears no resemblance to another hawk. It must have arrived a long time ago to have changed so much. But how long ago was that?⏎The answers have come recently by a
-
An example of such an occurrence is an earthquake. Serious interventions ought to be applied in a case such a tragedy occurs. Japan is one of the countries where earthquakes occur. According to Reid (2019), earthquakes in Japan happen around one
-
keeps your nitric oxide levels high. Which means closing tanning beds and beaches is actually horrible for heart health.⏎One of the reasons many think that viruses spread so rapidly during the winter and slow during the summer is the amount of UV radiation we
-
order to use natural uranium to fuel a reactor, it has to be extremely stingy with neutrons. The coolant moderator combination had to be one that did not absorb neutrons to any great extent. To make natural uranium “burn” every available
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
- L18:623 strongly suppresses
- L15:2176 excites
- L18:2077 excites
- L17:1688 excites
- L16:1577 suppresses
- L16:1789 excites
- L18:940 excites
- L13:2109 excites
- L16:1273 suppresses
- L18:1309 excites
- L18:1870 excites
- L14:1103 excites
- L17:1318 excites
- L18:2287 excites
- L14:1875 excites
- L11:922 excites
- L17:2292 suppresses
- L16:2362 suppresses
- L17:1921 suppresses
- L9:292 excites unclear: weak scattered tokens
- L16:2647 excites
- L18:2158 weakly suppresses
- L16:1720 weakly suppresses
- L9:1603 weakly suppresses
Feeds later layers
- L23:530 strongly suppresses
- L27:545 strongly suppresses
- L23:538 suppresses
- L21:542 suppresses
- L26:670 suppresses
- L20:2030 suppresses
- L27:620 suppresses
- L24:915 suppresses
- L21:699 suppresses
- L22:867 suppresses
- L20:1773 suppresses
- L20:1733 suppresses
- L20:1919 suppresses
- L27:755 suppresses
- L23:283 suppresses
- L24:1052 suppresses
- L23:2373 suppresses
- L24:1197 suppresses
- L20:914 suppresses
- L26:2494 suppresses
- L20:733 suppresses
- L20:851 suppresses
- L20:2341 suppresses
- L23:143 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
- ために
- _|
- |_
- ~
- '_
- theano
- 냥
- ~~
- voices
- //{⏎
Suppresses
- 很深
- 非常
- very
- 很大
- 较多
- Very
- .large
- 很
- 大量
- 很长