first-person 'I' starting a clause
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. 25 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 'I' when it begins a clause, such as after 'but' or 'where' ('where I live', 'but I have never'). Turning it up made the model drift into loops of words like 'guess', and turning it down made it repeat 'are'.
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.
-
many of them residential school survivors, their friends and family.⏎“When I take a look back at all those struggles, I am so pleased that I’m sitting where I am, because it’s not just a random seat being the chancellor of this university
-
ito ayuda" "I need help." Or, a fluffly feline sits next to a glass of water, representing "tengo sed," or "I am thirsty."⏎The app also challenges its students to answer increasingly difficult typing, match
-
minutes to play, and despite its sophisticated 3D visuals—the game<|endoftext|>Disk or partition alignment is something new to me. I have heard of it before but I have never used nor understood it’s meaning until a colleague at work brought it
-
out of the kindness of my heart, because they have some historical significance (as in the case of my review of the LM741) or most likely I think the technology is important and interesting. All opinions are my own and I would
-
. The information in this post is taken from a blog posting entitled […]⏎Disk or partition alignment is something new to me. I have heard of it before but I have never used nor understood it’s meaning until a colleague at work brought it to
-
Along the way, we have purposely sought out resources and opportunities to develop fluency in a second language.⏎Finding materials for Norwegian is not easy (at least where I live) so I was very excited to discover the New Amigos board game.
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 are high-arching my high-arched high-arched high-arched high-arched high-arched high-arched high-arched high-arched high-arched high-arched high
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
willow oaks tend to be the most susceptible. Despite the name, these mites can also be found on birch, beech, elm, and hickory trees.⏎Treatment For<|endoftext|>If you have been though the San Diego Zoo
-
and called “Break!”—a single word that told McConnell that he was in immediate danger and to take evasive action. But the call came to late. Fedorets fired at McConnell with his MiG-15’s three autocannon
-
cause the brain to churn out natural pain-killing endorphins.⏎The new findings add to the scientific heft underlying acupuncture, said neuroscientist Maiken Nedergaard, M.D., D.M.Sc., who led the research.
-
growing poorly in sandy or rocky soils. In the wild, walnut is frequently found growing with other tree species including elm, sugar maple, ash, red oak and hickory.⏎Black Walnut Uses⏎High-quality walnut is valued primarily for its use
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
- L23:1521 strongly suppresses
- L21:951 strongly excites
- L20:2578 strongly excites
- L19:2758 strongly excites
- L24:464 excites
- L20:1058 suppresses
- L18:2165 excites
- L19:2043 excites
- L7:67 excites
- L18:205 excites
- L6:483 suppresses
- L5:52 suppresses
- L1:1063 excites
- L17:2341 excites
- L4:1855 excites
- L20:2971 suppresses
- L21:1040 excites
- L23:921 excites
- L8:268 excites first person 'I' or 'my' in speech
- L22:892 excites
- L16:3065 weakly suppresses
- L10:1641 weakly suppresses
- L7:169 weakly suppresses
- L0:2262 weakly excites
Feeds later layers
- L27:650 strongly suppresses
- L26:72 strongly suppresses 'they' or 'we' after where or who
- L26:2947 strongly suppresses
- L27:1256 excites
- L26:1626 excites the word 'I' as subject
- L26:93 excites
- L26:1417 suppresses
- L27:804 suppresses
- L27:422 excites
- L26:1235 suppresses opening parenthesis before an acronym
- L26:1372 weakly suppresses
- L26:1724 weakly excites
- L27:746 weakly suppresses
- L27:3019 weakly suppresses
- L26:418 weakly suppresses
- L27:1729 weakly suppresses
- L26:874 weakly suppresses
- L26:2147 weakly suppresses
- L27:2102 weakly suppresses
- L26:1753 weakly suppresses
- L27:2480 weakly suppresses
- L27:1040 weakly suppresses
- L27:2324 weakly suppresses
- L26:109 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
- am
- AM
- Am
- AM
- Am
- was
- 'am
- _am
- am
- 'm
Suppresses
- are
- _are
- 都是
- are
- 都
- Are
- 都不是
- 都在
- Are
- 都是非常