'wide' in 'a wide range of'
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. 16 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 'wide' in phrases like 'a wide range' or 'a wide variety', and on 'hundreds' in 'hundreds of studies'. It turns off on words like 'sooner' in 'sooner or later'. Turning it up did not produce text that clearly showed this idea.
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.
-
a Safe Hospital?⏎When we prevent healthcare workers from facing a wide range of hazards on the job, including needle stick injuries, back injuries, latex
-
well as the symptoms and treatments associated with them⏎Understanding the Types of Broken Bones⏎The characteristics of broken bones can vary greatly. This means that there’s a wide range of
-
is increasing with each year. Unfortunately, a lot of that may include television shows, movies, video games, or YouTube videos that include violence.⏎There have been hundreds of research studies examining the effects of television violence on children and teenagers. The results
-
Grass Valley garden. Singer’s extensive experience gardening in the Sierra foothills provided the basis to formulate lists of plants deer avoid in our local growing areas.⏎Whether our wide-eyed friends devour prized roses or scrape the bark off new trees, deer can indeed
-
to type with hands-on advice for identifying the mystery interlopers in your borders, and information about their height, spread, flowering period and toxicity. Discover a wide array of British garden plants, with practical tips for their care and
-
10% flavan<|endoftext|>The eyes are the windows to the soul, or so goes the proverb. As humans, we use our eyes to express a wide variety of emotions, often without even realizing we’re doing
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 question to the next level (QNT]S] is to ask the question that is a question of * * * * * * * * * * * * * * * * * * * * * * * * *
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
worth considering at this time because, sooner or later, similar fights will break out in other areas. In the U.S
-
and made by consensus. They appreciate the discussion itself<|endoftext|>How to Unzip .tgz File Using the Terminal?⏎If you are using open-source packages, then sooner or later, you will encounter the .tgz or .tar.gz files.
-
1920 further asks the year of naturalization. Indirectly,⏎the 1820 and 1830 census provide a clue with the question "number of⏎foreigners in each household not naturalized."⏎
-
education. It was pointed out that boundaries and divisions are necessary to make sense of the complex world we live
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
- L15:1587 strongly excites
- L15:1900 strongly suppresses
- L15:1404 strongly suppresses
- L15:792 strongly suppresses
- L15:381 strongly excites
- L15:435 strongly excites
- L15:1616 strongly excites word fragments mid-word
- L15:1135 strongly suppresses
- L15:222 strongly excites
- L15:201 strongly excites
- L15:451 strongly excites
- L15:787 strongly excites
- L15:2803 strongly suppresses
- L15:1164 strongly excites
- L15:12 suppresses
- L14:126 excites
- L14:1090 excites
- L15:966 suppresses
- L14:169 excites
- L15:1043 suppresses
- L14:871 excites
- L15:1300 excites
- L15:992 excites
- L15:227 excites
Feeds later layers
- L18:2208 excites
- L19:1258 excites
- L22:849 suppresses
- L20:1163 excites
- L22:2961 excites
- L27:1731 suppresses
- L19:2839 suppresses
- L18:123 suppresses
- L26:154 excites
- L19:2889 weakly suppresses
- L26:1775 weakly suppresses
- L26:2854 weakly excites
- L20:2805 weakly suppresses
- L17:1971 weakly excites
- L19:1070 weakly suppresses
- L26:1430 weakly suppresses
- L26:192 weakly excites
- L26:1266 weakly suppresses
- L18:866 weakly suppresses
- L26:104 weakly suppresses
- L21:2548 weakly excites
- L26:2560 weakly excites
- L26:1979 weakly excites
- L26:2635 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
- лас
- בל
- Forbes
- Pruitt
- Clin
- kald
- oeff
- Freed
- ある
- 불구하고
Suppresses
- yx
- //**⏎
- y
- ism
- aat
- ioned
- qualifiers
- itives
- ivity
- !(: