split-off 's' starting a word
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. 0 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 a single 's' that starts a word after a bracket or a break, such as the 's' in '(saying' or '[s]aw', where the word has been split. Its negative side is the 's' at the end of a plural, such as 'Bottoms' or 'POPs'. Turning it up or down did not give a clear change in the model's writing.
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.
-
ab), a teacher from the fabled land of Olmo-lungring (‘Ol-mo lung-ring) on the eastern edge of Tagzig (sTag-gzig), who brought it to Zhang-zhung (Zhang-zhung
-
, as it was not possible to go across country,⏎equipment had to be man-handled from this point. "B" and "C" Batteries⏎saw plenty of action and subdued several machine gun nests that were holding⏎up the
-
But that doesn't mean you can't teach your 2-year-old to respect others. Begin with the basics (saying "please" and "thank you"), then try these six other strategies.
-
baobab, cowpeas, amaranth, vine spinach, Ethiopian kale, pumpkin leaves, spider plant and hibiscus.⏎Kale (sukuma wiki) is now the most common vegetable in Kenya.<|endoftext|>By Guest
-
ENTAL CELEBRATION OF THE PASCHAL MYSTERY⏎The catechesis of the liturgy entails first of all an understanding of the⏎sacramental economy (Chapter One). In this light, the innovation of its
-
placed there for fire protection) provided the source of water. Using three hoses at once, the pump sent the three streams of water over the United States Hotel (southeast corner
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 number of the number of the number of the number of the number of the number of the number of the number of the numbers. So, it's, it's, it's, it's, it's, it's, it
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
how is it still living if it’s gone?⏎It was home to teachers, farmers, horsemen, and Kentucky civil rights leader Jesse Bottoms.⏎Located off US 60 and the Bluegrass Parkway, Huntertown was
-
codeine. Examples of POPs are mercury, fire retardants (polybrominated diphenyl ether or
-
you<|endoftext|>Today, existing insulin pumps are about the size of a pager. The new ST- enabled Debiotech miniaturized MEMS device is about one quarter the size of these existing pumps and can be worn as
-
.aei.org/publication/the-omnivores-delusion-against-the-agri-intellectuals/⏎Leave a ReplyWant to join the discussion?⏎Feel free to contribute!<|endoftext|>
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
None: this is the first layer.
Feeds later layers
- L5:2594 weakly suppresses
- L14:2556 weakly excites
- L26:485 weakly excites
- L14:497 weakly excites
- L1:1575 weakly suppresses
- L5:2717 weakly excites
- L17:2736 weakly excites
- L8:1260 weakly suppresses
- L24:2544 weakly suppresses
- L7:1803 weakly suppresses
- L27:2428 weakly suppresses
- L7:2896 weakly suppresses
- L14:1735 weakly suppresses
- L19:1707 weakly excites
- L2:1683 weakly excites
- L6:248 weakly suppresses
- L20:1149 weakly excites
- L6:2534 weakly excites
- L19:146 weakly suppresses
- L9:720 weakly suppresses
- L17:1335 weakly excites
- L19:249 weakly suppresses
- L6:957 weakly excites
- L18:681 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
- ">-->⏎
- ounded
- riendly
- offee
- oom
- amble
- Physical
- iated
- physical
- PUTE
Suppresses
- kin
- recomm
- ść
- 缘
- LEY
- ELY
- adh
- 库
- 牵
- ker