'about' in 'it's not about' phrasing
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. 1 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 'about' in phrases like 'it's all about', 'it's not just about' or 'it isn't about', where the writer says what something is really concerned with. Turning it up did not clearly change the model's output.
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.
-
become apparent. It allows you to evaluate your behavior with food, which can help you identify changes that you can make to implement a healthier lifestyle.⏎It’s all about listening to your body’s cues. This allows you to realize when you’re hungry
-
and environmentally responsible use of Information Technology resources within an organization. It is not just about reducing the corporate carbon footprint; the goal of Green IT is also to significantly reduce
-
’m talking about is helping kids apply knowledge, think critically, imagine, be creative, be confident about being lifelong learners. It’s not simply about memorising facts and being able to recite that which one has memorised;
-
the current is bringing something, it’s all about experiences. Nature is everywhere. Birds, fish, animals and all kinds of trees and plants. It’s all about direction and distance and no choice is wrong.<|endoftext|>Prognosis and survival for
-
Health Disparities workgroup of Facing Race = Embracing Equity (FR=EE) is working to improve the doctor-patient relationship.⏎"It's not about, 'You're racist,' " said Candice Lucas, director of Community Health
-
: French History: Biographies<|endoftext|>White balance isn’t about the color of the sun – it’s about the color of the light. The
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 working on this problem, and I am not sure of the answer. So I need to think about it again. I need to think about it again. So I have been thinking about this problem, and I am not sure
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and Denmark offered him large honorary pensions which he modestly declined. At the urging of fellow scientists he attempted to form an imperial academy of science, but was thwarted by political enemies of the Jesuits. He did succeed in publishing a very timely
-
his compatriot Sutan Sjahrir sought to join forces with other nationalist leaders, including Sukarno. The organizational efforts of these men were thwarted by the repressive policies of the colonial state. First Sukarno was arrested
-
bigger killer: infectious disease.⏎"When a child is diagnosed, parents ask: Why is this in our genes?" said Butte, assistant professor of pediatric cancer biology and a bio
-
to distinguish themselves as cancer stem cells. In the K-RAS model, all tumor cells were equally likely to propagate tumors<|endoftext|>Pharmacists are caring for more individuals of diverse age, gender, race, ethnicity, socioeconomic status,
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
- L0:918 weakly excites
- L0:558 weakly suppresses inside all-caps abbreviations
- L0:315 weakly suppresses
- L0:3068 weakly excites
- L0:559 weakly suppresses
- L0:1045 weakly excites
- L0:1301 weakly suppresses
- L0:1955 weakly suppresses
- L0:1224 weakly suppresses
- L0:873 weakly suppresses
- L0:178 weakly suppresses
- L0:1240 weakly suppresses
- L0:1727 weakly excites
- L0:483 weakly excites
- L0:249 weakly excites ordinal suffix "nd" after numbers
- L0:1758 weakly excites
- L0:532 weakly excites
- L0:2283 weakly excites
- L0:2061 weakly excites
- L0:2323 weakly excites
- L0:1555 weakly suppresses
- L0:1556 weakly excites
- L0:1114 weakly excites
- L0:513 weakly suppresses
Feeds later layers
- L3:309 suppresses
- L2:584 weakly suppresses
- L3:2608 weakly suppresses
- L5:1839 weakly excites
- L6:819 weakly excites
- L18:170 weakly suppresses
- L15:1551 weakly excites
- L20:1628 weakly excites
- L2:1409 weakly suppresses
- L15:414 weakly excites
- L17:1847 weakly excites
- L15:1847 weakly excites
- L13:1851 weakly excites
- L9:115 weakly excites
- L25:1943 weakly suppresses
- L2:740 weakly excites
- L17:2683 weakly excites
- L6:165 weakly suppresses
- L4:112 weakly suppresses
- L4:616 weakly suppresses
- L16:2202 weakly excites
- L12:1071 weakly excites
- L18:1913 weakly suppresses
- L10:1320 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
- oton
- ewis
- ativo
- 踏
- 对他们
- alars
- iversary
- tail
- setAddress
- 对未来
Suppresses
- ings
- ῇ
- clam
- egan
- чки
- чив
- ewidth
- .Deep
- êu
- jeta