"way of" before life, looking, eating
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. 6 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
The neuron lights up on the word "of" right after phrases like "way of", "ways of" or "mode of", where a noun like life, looking or eating is about to follow. When it is turned up, the model drifts into vague, abstract talk about "the world of being" and repeats "of" a lot. Turned down, it writes more concrete, everyday sentences.
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.
-
ahili is spoken by many.⏎Kenyan Tribes & People⏎Kenya is built on traditional tribal cultures and each tribe contributes to the unique lifestyle and way of life so many are intrigued by. There are 42 tribes across Kenya,
-
constitution (Article 2), Islam is the state religion. By the constitution's Article 31, it is the country's duty to foster the Islamic way of life. By Article 33, it is the country's duty to discourage
-
the one hand, given the scale and scope of the threats that they face with regard to climate change – including specific threats to their livelihoods, cultures and ways of life – their situation is different from that of other groups and from that of the
-
Mindset⏎One thing that should be stated is that inventors do not always think like the people who follow their inventions. They have their own mindset and way of looking at things. This should be obvious, because they are seeing *everything*
-
is packed with nutrients, light on calories, and high in fiber. In a nation known for massive portion sizes and ready access to highly processed foods, this way of eating should be a welcome change. But a closer looks reveals a more complicated,
-
th century. He also helped start the cognitive revolution in psychology through his review of B. F. Skinner's Verbal Behavior. He challenged the behaviorist way of looking at behavior and language. This was the main approach used in the 1
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 have been thinking about the history of the people in the area. The people will say that I have been thinking about the historical background of the area. I have been thinking about the past. This is a case where the people might be
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
of government policy.⏎The goal is "to narrow as soon as possible the gap in economic development between Tibet and other areas of the nation" (White Paper on Tibet, Sep 1992). Chen Kuiyuan, a Chinese
-
by the Chinese. What was Tibet's homeland will become China's homeland. Beijing has finally admitted that the population transfer process goes on. (Third Work Forum on Tibet, July 1994). A wave of resettlement became apparent in
-
the locking up of perhaps 1 million Uyghurs, a mostly Muslim ethnic minority, in China’s at times violence-torn far western region of Xinjiang.⏎All anniversaries in China are sensitive, with the party eager to control
-
(International Campaign for Tibet 2/11/94). These settlers recieve subsidies and other
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
- L3:1537 weakly excites
- L4:37 weakly suppresses
- L3:2987 weakly excites
- L1:2612 weakly excites
- L3:2721 weakly excites
- L5:2117 weakly excites
- L4:2699 weakly excites
- L5:2061 weakly excites
- L2:1660 weakly excites
- L3:2886 weakly suppresses
- L4:144 weakly excites
- L1:979 weakly suppresses
- L4:3047 weakly excites
- L4:1537 weakly excites
- L4:1498 weakly excites
- L3:885 weakly suppresses
- L1:2041 weakly excites
- L0:2963 weakly excites
- L4:1110 weakly excites
- L4:784 weakly suppresses
- L4:159 weakly excites
- L3:3054 weakly excites
- L3:692 weakly excites
- L2:1523 weakly suppresses
Feeds later layers
- L8:202 excites
- L7:2795 suppresses
- L11:1392 weakly excites 'of' in fixed technical phrases
- L23:2213 weakly suppresses
- L10:56 weakly suppresses
- L10:1836 weakly suppresses
- L7:2607 weakly excites
- L17:1015 weakly excites
- L25:1753 weakly suppresses
- L26:2257 weakly suppresses
- L13:1335 weakly excites 'of' inside fixed noun phrases
- L23:647 weakly suppresses
- L22:2130 weakly excites
- L15:2276 weakly suppresses
- L9:849 weakly suppresses
- L25:968 weakly excites
- L24:1382 weakly excites
- L18:43 weakly excites
- L10:239 weakly excites
- L22:1148 weakly suppresses
- L14:2397 weakly suppresses
- L21:131 weakly excites
- L17:1771 weakly excites
- L9:116 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
- loom
- Document
- 舸
- ownership
- �
- كوك
- motors
- TokenType
- fen
- หมาย
Suppresses
- 浴室
- arie
- остоя
- 因地
- urface
- 这种事情
- icina
- uffer
- handling
- пров