'to' after a need or purpose
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. 21 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 responds to the word 'to' when it introduces a purpose or need, as in 'depend on sea ice to live' or 'need water to drink'. Turning it up made the model repeat 'while' and similar small connecting words, so the link to that theme is weak.
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.
-
on Earth.⏎Why Does Melting Ice Matter?⏎This rise in temperature dramatically throws off the balance of the Antarctic ecosystem. Many organisms in Antarctica depend on sea ice to live. Plants and shrimplike crustaceans called krill grow on
-
you eat after a certain time, you are guaranteed to gain weight. Based on your body type and height, your body needs a specific number of calories in order to maintain your present weight. If you consume all of those calories close to bedtime,
-
Just a few months ago, wetlands were brimming with water, but now many are just parched expanses of cracked mud.⏎“Birds need water to drink every day, and when the wetlands dry out, they need to find
-
photosynthetic organisms. Hydrothermal vents, submarine hot springs, methane cold seeps, and whale falls rely only on chemical energy and chemosynthetic organisms to support life.⏎Fundamental Concept 5h.⏎Tides, waves and pred
-
might be cut down a few years later<|endoftext|>Hand-coloured lithograph entitled; 'Dampfmaschinen Pferd worauf man in einer stunde von Paris nach Petersburg reiten kann' ('Steam engine horse on which one can
-
much healthier.⏎So says endocrinologist Andrew Greenberg, director of the Obesity Metabolism Laboratory at Tufts University in Boston.⏎Your body needs fat cells to be healthy, but in obese individuals when fat cells get very big, those cells
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 most important and always **perce + *ce* + *se** * + *se* + *ce** *** + *se** * + *ce** * + *ce** * + *** + *se
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
and air conditioning systems, it also means that pollutants that originate indoors are getting trapped inside.⏎“People normally<|endoftext|>Spring Dawn by Meng Haoran 春眠不觉晓 处处闻啼茑 夜来风雨声 花落
-
221BC). Unlike most Chinese idioms, 毋忘在莒 strangely has actually two stories of origin, which gave this idiom two
-
a matrix⏎Find nth maximum⏎Matlab Basics II - Create a vector with a repeated entry⏎Create sequnce 1 4 9 16 25.........⏎Basics
-
for the entire class.⏎PC: Windows XP and up⏎Mac: OS X (10.1) and up⏎Resolution: 800 x 600 or higher<|endoftext|>In response to the theme of 201
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
- L15:184 excites
- L13:44 suppresses inside web addresses and citations
- L5:7 suppresses
- L11:223 weakly excites
- L10:155 weakly excites
- L18:2520 weakly excites
- L18:409 weakly excites
- L12:158 weakly suppresses
- L19:2264 weakly suppresses
- L18:274 weakly excites
- L18:2789 weakly suppresses
- L16:899 weakly excites
- L4:852 weakly suppresses
- L19:154 weakly suppresses
- L20:1638 weakly suppresses
- L18:178 weakly excites
- L10:1715 weakly suppresses
- L20:1060 weakly suppresses
- L15:610 weakly excites
- L2:75 weakly suppresses
- L20:2302 weakly excites
- L19:2388 weakly suppresses
- L20:2046 weakly excites
- L16:1479 weakly excites
Feeds later layers
- L23:1127 excites
- L22:1226 excites
- L22:2019 weakly excites
- L24:2708 weakly suppresses
- L27:170 weakly suppresses unclear: spaces in mixed-language text
- L27:760 weakly excites
- L23:2122 weakly excites
- L27:1358 weakly excites
- L23:1213 weakly excites
- L25:1381 weakly excites
- L22:1865 weakly suppresses
- L23:1926 weakly suppresses
- L23:1344 weakly suppresses
- L22:1316 weakly suppresses
- L25:672 weakly excites
- L26:637 weakly suppresses
- L22:267 weakly suppresses
- L24:2978 weakly excites
- L22:1694 weakly suppresses
- L23:769 weakly suppresses
- L23:2062 weakly excites
- L22:131 weakly excites
- L22:2937 weakly excites
- L23:278 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
- 而
- 前行
- 而出
- 前进
- 起
- while
- 成长
- while
- 出场
- 行事
Suppresses
- люб
- Stores
- ments
- iture
- $body
- jen
- igure
- sez
- Fixes
- propri