'words' in 'in other words'
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. 11 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 lights up on the word "words" in the phrase "In other words", where a writer restates an idea more plainly. Turning it up gave replies that drifted into repetitive lists, which does not show the phrase's meaning.
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.
-
not only includes controlling a specific cell, but it also allows for creating a secondary drop-down list that is completely dependent on the first or primary list. In other words, once the first list is displayed, and a parameter chosen, then a subset
-
many skeptics of Christianity take a simple approach to reading the Bible. They treat the entire Bible and all of its contents as a moral command textbook. In other words, every single sentence is to be read with an eye toward what moral behavior the
-
recognition that a sedentary lifestyle is being driven, at least in part, by environmental factors that affect individuals' physical activity choices and health behaviours. In other words, the environments in which we live, and with which we interact, have become
-
learning experience for students. As a community college, it is understood that Kirtland maintains somewhat of an obligation to connect the classroom to the community. In other words, service to the community performed by students is consistent with community college tradition. This
-
these rankings were used to grant a standard reward to a top group of teachers, not to make fine distinctions in the amount of compensation that teachers received. In other words, the results of value-added analysis, when examined over a period of two or
-
about the prospect, and the board as a whole made it clear that their tentative decision does not mean a regulation of this nature would ultimately be approved. In other words, as of the date of this writing, it is still illegal to kill g
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: the fact that, as I have been told, and I have been told by the judge, and I have been told by the judge, and I have to be. So I have to be. This is a really important thing to be
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
as being a List of Users.⏎Sometimes, it’s also handy to define a list of multiple Entities and Structures.⏎Until 184.108.40.2
-
uler or calipers with mm divisions⏎- Paper towel⏎- Microscope slide and cover slip⏎AQA: Topic 184.108.40.206 and 220.1<|endoftext|>Re export
-
Volume (Year): 42 (2011)⏎Issue (Month): 2 (June)⏎|Contact details of provider:|| Web page: http://www.tandfonline.com/VECE20|⏎|Order Information
-
to agriculture.⏎For more information, please contact:⏎Dr. Norihito Nakamichi⏎Biodynamics Research Team⏎RIKEN Plant Science Center⏎Tel: +81
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
- L5:134 suppresses
- L9:396 excites colon after a metadata label
- L10:572 excites
- L6:1285 excites
- L4:946 suppresses
- L9:2604 weakly excites
- L9:2159 weakly excites
- L5:1526 weakly suppresses
- L5:1847 weakly suppresses
- L9:2590 weakly suppresses
- L6:1016 weakly suppresses
- L9:97 weakly excites
- L4:193 weakly excites
- L7:2354 weakly excites
- L6:74 weakly excites
- L9:84 weakly suppresses
- L8:749 weakly suppresses
- L0:3037 weakly suppresses
- L4:51 weakly suppresses
- L8:747 weakly excites
- L5:2364 weakly suppresses
- L9:797 weakly excites
- L2:2826 weakly excites
- L1:1931 weakly suppresses
Feeds later layers
- L17:2684 excites
- L14:2774 excites
- L16:2241 excites
- L17:2365 excites
- L15:598 suppresses
- L17:806 excites
- L13:895 excites
- L15:38 weakly excites
- L13:584 weakly suppresses unit conversions in parentheses
- L14:2156 weakly excites opening parenthesis after a term
- L20:1772 weakly excites
- L17:1229 weakly excites
- L12:1506 weakly excites
- L15:261 weakly excites
- L12:1820 weakly suppresses
- L12:1837 weakly excites
- L14:2610 weakly excites
- L17:2605 weakly suppresses
- L16:1394 weakly suppresses
- L12:959 weakly suppresses
- L17:2364 weakly excites
- L17:1988 weakly excites
- L16:656 weakly suppresses
- L13:626 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
- dating
- antar
- doi
- amt
- śmy
- posite
- REST
- ressing
- \":{\"
- uspendLayout
Suppresses
- 썸
- 然后
- untu
- ))]⏎
- 슈
- 페
- atinum
- 冠军
- тогда
- iteration