'come' in phrases like 'come across'
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. 23 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 'come' when it is part of a phrase, such as 'come across', 'come up with' or 'come to'. It is switched off by 'came' and 'person' in some cases. Turning it up gave replies that were loosely about thinking or coming to a conclusion, but the link is only partly clear.
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.
-
hand to genitals), and more! Because there isn’t just one type of sex, there are different options for safer sex. These are some terms that you might come across when we talk about safer sex.⏎Antiretroviral
-
out how you came to be on this earth and who lived before you in the family tree.⏎The most common type of DNA testing for genealogy that you’ll come across is an autosomal DNA test. Not only do they tell you your DNA
-
be difficult to come up with a great melody starting only with lyrics.⏎Maintain a collection of lyr
-
we don’t keep the end in mind. For example, a teen working to save $50 to have spending money for entertainment will more easily be able to come up with a plan than if they were trying to save up $5,0
-
there was no consensus on the definition of a microservice and what it actually means. It is quite a buzzword these days but none of the panelists could come to a common understanding. This raised a few questions in my head. Disappointed
-
asking them to download the free Aurasma app and bring their device. We talk about the importance of "depth" in education. I'm sure you'll come up with lots of ideas of your own. A. Media:
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: walking walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk walk
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
’s This?⏎The game went to penalty kicks and, right when Messi was about to take one, Uncle Simon, dressed in his usual beehive wig, came in with a tray of canapés and asked, ‘Who’s playing?’
-
problem solver and go out of your way to help co-workers. You do your job and make it all look easy. Despite your dedication, one day a new person is hired. They have the same education, experience, and skills that you have
-
in the Chaos⏎The perils of plastics in the ocean are threatening to the ecosystem, but all is not gloomy. This is where BLUE BEAUTY comes to the rescue. It’s a sub-section of clean or green beauty but domin
-
these machines that can capture an underlying spirit, or an underlying set of principles, or an underlying way of looking at the world so that then when the next Aristotle comes around?" Jobs said. "Maybe if he carries around one these machines with him
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
- L21:2728 suppresses
- L12:1826 weakly suppresses
- L20:2514 weakly suppresses
- L15:1019 weakly suppresses
- L17:217 weakly suppresses
- L12:1926 weakly excites
- L16:1853 weakly suppresses
- L6:1205 weakly suppresses
- L20:380 weakly excites
- L21:1221 weakly suppresses
- L0:2541 weakly suppresses
- L7:1851 weakly suppresses
- L5:2324 weakly excites
- L18:1407 weakly excites
- L22:151 weakly suppresses
- L4:360 weakly suppresses
- L7:2166 weakly excites
- L6:465 weakly suppresses
- L22:2909 weakly excites
- L17:1469 weakly excites
- L15:2109 weakly excites
- L3:90 weakly suppresses
- L0:450 weakly suppresses
- L15:1751 weakly suppresses
Feeds later layers
- L24:1219 suppresses
- L26:1965 suppresses
- L25:1027 weakly suppresses
- L25:194 weakly excites
- L24:1622 weakly excites
- L26:3039 weakly suppresses
- L27:824 weakly suppresses
- L25:2115 weakly suppresses
- L27:1008 weakly excites
- L24:691 weakly suppresses
- L24:411 weakly excites
- L27:1050 weakly excites
- L24:1301 weakly suppresses
- L27:2634 weakly excites
- L26:1104 weakly excites
- L25:2236 weakly suppresses
- L27:2474 weakly suppresses
- L27:2129 weakly excites
- L27:235 weakly excites
- L25:1943 weakly suppresses
- L27:1381 weakly excites
- L27:1526 weakly suppresses
- L24:876 weakly suppresses
- L25:754 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
- up
- _Config
- Away
- Away
- up
- لين
- away
- -up
- Guth
- away
Suppresses
- walk
- walk
- 走了
- walks
- walked
- 走
- walking
- entar
- ⾛
- 步