'important' and degree words after 'how'
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. 10 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. 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 words that judge how much something matters or how strong it is, especially 'important' right after 'how' (as in 'how important is...'). When it was turned up, the model's replies were unrelated and a bit scrambled, so the steering tells us little about its role.
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.
-
something new: If you baby-sit for your grandchildren, how about walking to the park instead of playing video games?⏎- Rethink your priorities: How important is an entire afternoon of TV? How about a walk after lunch instead?⏎-
-
– coal, shale gas – affect climate change. Thinking long-term is not about pin-pointing what will happen in a given year. It is about understanding how important trends work over time.⏎The same is true of the budget. A longer window
-
on governments and businesses - and even individuals - to carefully study what kind of hazards could trigger a disaster, to identify what is exposed to these threats, how sensitive an area is to these hazards and what would be the consequences of a storm like Sandy
-
such as diet and exercise can have a deep positive impact on our health and lower high cholesterol levels, it turns out that genetics also plays a crucial part in how high our cholesterol may end up
-
due to lockdown and less travel. If we still use our cars less after the pandemic, this could continue to make a huge difference.⏎But, how much of a difference are we talking?⏎Well,
-
a tool for assessing the democratic health of European countries across many different dimensions. That includes not just formal dimensions of democracy but also more everyday features of democracy – how important democratic principles and practices are to the cultures of workplaces, to people’s community life
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: what do do I should do do I can have a child I have will be have a child I have will be have a baby I will be have a child I have will be a child is also a place is a person is a child
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, now you will find below the Solutions of Class 10 Science Chapter 4 Environmental management questions answers. From the subject English of Maharashtra State Board.⏎Below you can see
-
information about children's rights available in one place is to build a clearer picture of some of the repeated violations of children's rights in a given country. The eventual goal will be to try to match these violations with possible avenues of redress.⏎We
-
1, 2009⏎Computer science focuses on aspects of the computer field that have a well-defined set of abstract concepts and principles. The objective of a concentration in computer science is to demonstrate strong comprehension of those abstract concepts and principles
-
how true is that! There are more job opportunities in Analytics and Optimization than there were last year and many IT professionals are prepared to invest time and money for the
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
- L9:2773 strongly suppresses
- L2:89 strongly suppresses
- L7:403 excites
- L8:511 excites
- L2:180 suppresses 'is' after focus or goal nouns
- L5:1038 suppresses
- L4:670 excites
- L6:217 excites
- L9:756 excites
- L5:1938 suppresses
- L5:224 suppresses
- L8:1149 excites
- L5:1717 excites
- L5:99 excites
- L8:174 excites
- L9:1258 excites
- L7:84 suppresses
- L5:2147 excites
- L2:638 excites
- L2:578 excites
- L9:2303 suppresses
- L5:395 suppresses
- L9:2357 suppresses
- L8:1931 excites
Feeds later layers
- L15:304 strongly excites
- L13:1129 strongly suppresses
- L11:578 strongly suppresses
- L16:968 strongly excites
- L13:1107 strongly suppresses
- L15:681 strongly suppresses
- L18:857 strongly suppresses
- L15:2971 strongly suppresses
- L11:782 excites
- L13:403 excites
- L13:2672 suppresses
- L16:1223 suppresses
- L15:824 suppresses
- L23:220 suppresses
- L14:1326 suppresses
- L14:850 suppresses
- L25:9 excites
- L21:262 excites
- L12:869 suppresses
- L20:586 suppresses
- L14:1487 suppresses
- L16:1808 suppresses
- L18:907 suppresses
- L16:881 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
- ích
- bidden
- ーター
- ople
- 些什么
- 足够的
- ooth
- cht
- aster
- ewhere
Suppresses
- 是
- 是我
- 是他
- 是我的
- 是由
- 是没有
- 是不会
- adalah
- 留下了
- 能得到