'and' joining two clauses
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. 17 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
The neuron lights up on the word 'and' when it joins two full sentence-like parts, such as 'she'll never be as sociable as an extrovert and that's OK'. The opposite side fires on 'and' after a comma in a list, but turning the neuron up did not give a clear theme in the writing.
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.
-
Remember, however, to take your teen’s natural disposition into account. If she’s an introvert she’ll never be as sociable as an extrovert and that’s OK. More importantly, with a great attitude, she’ll attract like
-
is to drive with a BAC of .08 percent.⏎Blood alcohol concentration, or BAC, describes the amount of alcohol contained in an individual’s blood and is measured as weight per unit of volume. The measurement is usually
-
teaches its readers that, in life, we must make choices, and learn to live with and move on from said choices once there is no way back to them and life has moved on. I took the one less travelled by
-
by their own living experience. Adoptees generally live a regular life similar to someone who wasn’t adopted; however, they have experiences that are unique to being adopted and these experiences may have an impact on their lives at various times throughout their lives.
-
housings, etc. Any two different metals under water will cause a small electrical current that will steadily corrode one metal. Sacrificial anodes are the solution and we’ve been using them for decades, no, make that centuries now.⏎It
-
but to maintain enough energy in general. Think about it: an ultramarathon is any race that is longer than the traditional 26.2 mile marathon and can commonly reach distances of 100 miles or more! Even short runs
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: How to, in our society, that the way, that the world, that world is, but in our eyes, we still, we have, but the world, but world, but world, but world, but world, but world
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
experience, and seventy-six percent know of someone who has experienced seasonal depression.⏎(Male 2
-
keywords for each book and chapter.⏎If you think you should have access to this title, please contact your librarian.⏎To troubleshoot, please check our FAQs, and if you can't find the answer there, please contact us.<|endoftext|>Indigenous
-
refine, and establish clinical uses for liquid biopsy tests.⏎Different liquid biopsy<|endoftext|>Biology 1
-
copy, and fecal occult blood test.⏎- Cervical cancer: Regular Pap tests are
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
- L13:131 strongly excites 'and' joining parallel phrases
- L16:376 suppresses 'or' joining alternatives
- L10:74 excites
- L9:92 suppresses "and" or "or" joining two items
- L4:57 suppresses
- L5:548 excites
- L16:1669 excites
- L15:1443 excites
- L15:727 excites
- L1:1250 suppresses
- L16:2049 excites
- L10:1643 suppresses
- L14:1639 suppresses "and" joining two nouns
- L4:2270 weakly excites
- L13:1066 weakly excites
- L15:44 weakly excites
- L12:51 weakly excites
- L14:1237 weakly excites
- L14:801 weakly excites 'and' or 'or' joining list items
- L14:936 weakly excites 'and' or 'or' in a list of three or more
- L4:108 weakly excites "or" offering a choice in web text
- L0:313 weakly excites "and" joining paired nouns
- L0:2711 weakly suppresses
- L6:129 weakly suppresses
Feeds later layers
- L19:2493 excites
- L19:1245 suppresses comma opening a list of parallel items
- L19:2416 excites
- L18:765 excites
- L19:1262 suppresses
- L18:1047 suppresses
- L18:892 suppresses
- L19:979 excites
- L18:2516 suppresses
- L24:224 weakly suppresses
- L19:3015 weakly excites
- L19:3001 weakly excites
- L26:610 weakly suppresses
- L22:118 weakly excites
- L19:1506 weakly excites
- L20:658 weakly excites
- L18:2341 weakly suppresses comma before an added clause
- L19:577 weakly suppresses
- L18:174 weakly excites
- L21:478 weakly suppresses
- L20:852 weakly suppresses
- L19:291 weakly excites 'and' joining two full clauses
- L21:739 weakly suppresses 'both X and Y' pairing words
- L18:2222 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
- stand
- 掖
- 眼里
- "));⏎
- gang
- cek
- "]);⏎
- vested
- 全体员工
- 谈到
Suppresses
- orca
- ude
- icrobial
- .Consumer
- iders
- icerca
- sburg
- afari
- umeric
- .constructor