"or" offering a choice in web text
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. 4 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 "or" when it joins two options, like "Download or read it online" or "all or most of the teeth". Turning it up made the model fall into repeating boxed or template-like text, so the steering result does not clearly show 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.
-
measured unless subject to both ethical and commercial consideration.⏎Home page url⏎Download or read it online for free here:⏎by Flora Klickmann - G.
-
, as in any such course, is to cover the fundamental concepts of computer graphics, and the concentration is on graphics in three dimensions.⏎Home page url⏎Download or read it online for free here:⏎by Fränk Zimmer - Wolke Ver
-
Minecraft to product instant and awesome results. Craig Richardson is a trainee Computing and ICT teacher originally from the North East of England and working in East London.⏎Download or subscribe to this show at https://twit.tv/shows/triangulation.⏎
-
the percentage of infected participants that was<|endoftext|>If you are participating with a teacher, please enter your teacher's code here:⏎An email has been sent to you (or your parent if you are under 18) to get you signed up for
-
genealogical research - or rule out dead ends. Click on the link above to learn more, sign up for the project by filling out the form below, or contact firstname.lastname@example.org for more information. Updated June 25,
-
the walls of grocery stores and bus stands: “No Kidnapping. No Torture. No Extortion,” written by Bodoland National Council; “Do or Die for Bodoland,” by All Bodo Students Union. In a quiet Kok
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 is, "the question is, the question is...". So, I can say that, I can say that... and so. So, in the context of modern, and so. So. I think. So. I
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
postlude, or interlude, or any short instrumental work.⏎note2: Many symphonies are tonal works in
-
a simple and cheap way to integrate renewable energy into your home by turning your appliances into solar electric hybrids. Here is how it works. A solar panel (or any other renewable power source) charges a storage battery. A
-
to upload the pdf to Brightspace or Google Classroom (or any other platform you are using):⏎Have students record answers on top of the pdf⏎
-
may be used on other projects or donated to Habitat Restore.⏎- Transportation jobs - If you work for a trucking company, a city or state transportation department or any other organization that uses
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
- L1:1710 suppresses
- L3:1200 excites
- L0:1882 suppresses unclear: 'high tech' and mid-word pieces
- L3:1568 excites
- L3:2783 suppresses
- L2:70 weakly excites
- L3:115 weakly suppresses
- L3:250 weakly excites
- L3:1960 weakly suppresses
- L3:122 weakly excites
- L3:512 weakly suppresses
- L3:272 weakly excites
- L2:1494 weakly suppresses
- L2:748 weakly suppresses
- L2:2776 weakly suppresses
- L1:1404 weakly suppresses
- L1:1099 weakly excites
- L3:1169 weakly suppresses
- L3:2031 weakly excites
- L2:75 weakly suppresses
- L3:464 weakly excites
- L3:2346 weakly suppresses
- L0:873 weakly suppresses
- L3:2140 weakly suppresses
Feeds later layers
- L10:135 strongly excites 'or' in alternative requirements
- L18:2197 strongly excites
- L15:2935 excites
- L15:1907 excites the word 'or' between alternatives
- L9:92 excites "and" or "or" joining two items
- L6:2818 excites
- L8:2002 excites
- L5:872 excites 'or' introducing an alternate name
- L5:1203 excites
- L16:2049 excites
- L11:406 excites
- L8:1475 excites
- L5:880 excites
- L9:2010 excites
- L21:735 excites 'or' in ranges and alternatives
- L18:870 excites
- L7:836 excites
- L17:2671 excites the word matter in no matter phrases
- L19:1754 excites
- L8:562 excites
- L13:131 excites 'and' joining parallel phrases
- L13:553 excites
- L11:251 excites
- L6:1052 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
- phans
- ients
- 其它
- kest
- thon
- naments
- ifice
- atory
- lando
- else
Suppresses
- 这一点
- greso
- ei
- ually
- edly
- ッシ
- ocy
- gly
- UDGE
- hear