Neuron Explorer

Qwen3-0.6B

'but' in 'not only X, but also' contrasts

A neuron in layer 16 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the word 'but' when a sentence adds a contrast after a list or a 'not only' phrase, such as 'not only oceans, but also soil'. It is more negative on the word 'by' in phrases like 'burned by'. Turning it up gave wandering text with no clear shared theme.

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 activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    the bodies of people who don’t.⏎Blood sugar, or glucose, is the body’s main source of energy. This glucose comes from food—mostly carbohydrates, but occasionally

  2. #2 peak

    of algae-like plants called Diatoms.⏎Diatoms occur in virtually every environment that contains water. This includes not only oceans, seas, lakes and streams, but also soil. It is believed that 30 million years ago the diatoms

  3. #3 peak

    about the distribution of rock such as limestone, sandstone, coal and granite. Surficial material maps describe deposits that occur above bedrock. This includes soil, but also details up to several hundred feet of deeper unconsolidated material.⏎-

  4. #4 peak

    .htm ) BOYLE Last Name Meaning & Origin: A variant of O'BOYLE, from the Irish Ó BAOGHILL. Of uncertain derivation, but the Boyle last name is considered by most to be connected to the Irish geall

  5. #5 peak

    diet does focus on eating lots of plant-based foods, it does not necessarily mean you have to become a vegetarian or vegan. Meat and dairy are perfectly acceptable, but the idea

  6. #6 peak

    04 construction for the new Heilwood Company store had begun. It would be a two-story, wooden frame building and included not only a general store, but also the town post office. The store manager’s house was constructed directly below the

Steering

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.

turned down a lot
Prompt

Here is something I have been thinking about lately: the fact that a single one of the two the two the one is the one the one to be, the one to be that, the one to be, the one to that, the one to the one, the one to the one

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    version of bool-extras is 0.3.0 and is available on Hackage.⏎If you have cabal-install (which you really should have, by the way), the

  2. #2 peak

    of having succulents:⏎- Succulents absorb harmful chemicals from the<|endoftext|>"Everything we see hides another thing, we always want to see what is hidden by what we see." --Rene Magritte⏎D.B. Johnson writes and

  3. #3 peak

    , companies display total calorie burn statistics rather than net calorie burn numbers which help account for things such as your metabolism, so that you know how many calories are burned by that exercise in particular rather than just during that time period.⏎If you're walking

  4. #4 peak

    quite true. Even though they represent one family of plants, there are some distinctive features between the two. In this regard, it is necessary to indicate several criteria by which it is possible to determine the difference between these plants. In this article,

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L2:1855 excites
  2. L4:1319 excites the 'B' in vitamin B
  3. L1:4 excites
  4. L6:328 weakly suppresses
  5. L15:2115 weakly excites
  6. L7:520 weakly suppresses
  7. L10:201 weakly suppresses the letter B in vitamin names
  8. L9:2839 weakly excites
  9. L0:825 weakly suppresses
  10. L15:1837 weakly excites
  11. L5:589 weakly excites
  12. L6:1357 weakly excites
  13. L1:2928 weakly excites
  14. L7:566 weakly excites
  15. L12:606 weakly excites
  16. L9:1385 weakly suppresses
  17. L8:1187 weakly excites
  18. L11:288 weakly excites
  19. L6:2106 weakly suppresses
  20. L10:887 weakly suppresses the B in vitamin B names
  21. L15:695 weakly suppresses
  22. L13:1423 weakly suppresses
  23. L14:1273 weakly suppresses
  24. L2:1769 weakly excites

Feeds later layers

  1. L17:420 excites
  2. L18:404 excites 'get', 'go' and 'got' after a subject
  3. L17:230 excites
  4. L17:300 excites
  5. L19:526 excites
  6. L18:105 excites
  7. L18:936 excites
  8. L19:299 excites
  9. L17:1313 weakly excites
  10. L20:200 weakly excites
  11. L19:2795 weakly excites
  12. L20:1852 weakly excites
  13. L20:1624 weakly excites
  14. L18:198 weakly excites
  15. L19:221 weakly excites
  16. L17:594 weakly excites
  17. L17:509 weakly excites
  18. L17:241 weakly excites
  19. L18:1741 weakly excites
  20. L19:371 weakly excites
  21. L19:359 weakly excites
  22. L19:683 weakly excites
  23. L19:593 weakly excites
  24. L19:893 weakly excites

Activation histogram

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.

2 values between -4.89 and -4.651 values between -4.65 and -4.401 values between -4.40 and -4.161 values between -3.92 and -3.682 values between -3.68 and -3.442 values between -3.44 and -3.201 values between -2.96 and -2.714 values between -2.71 and -2.473 values between -2.47 and -2.235 values between -2.23 and -1.996 values between -1.99 and -1.7511 values between -1.75 and -1.5116 values between -1.51 and -1.2715 values between -1.27 and -1.0231 values between -1.02 and -0.7853 values between -0.78 and -0.54166 values between -0.54 and -0.30906 values between -0.30 and -0.062181 values between -0.06 and 0.18372 values between 0.18 and 0.4272 values between 0.42 and 0.6727 values between 0.67 and 0.918 values between 0.91 and 1.154 values between 1.15 and 1.395 values between 1.39 and 1.636 values between 1.63 and 1.871 values between 1.87 and 2.111 values between 2.11 and 2.362 values between 2.84 and 3.081 values between 3.56 and 3.801 values between 4.29 and 4.531 values between 4.53 and 4.77offnegativepositiveoften

Logit lens

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 layers instead.

Promotes

  1. irez
  2. chaft
  3. riel
  4. bler
  5. ynthesis
  6. amework
  7. setType
  8. setC
  9. nesty
  10. ufact

Suppresses

  1. 的就是
  2. ERR
  3. _None
  4. events
  5. hibited
  6. event
  7. course
  8. 喙
  9. stein
  10. None