Neuron Explorer

Qwen3-0.6B

opening quote marks and colons before quoted text

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

What this neuron does

This neuron fires on the quotation mark, colon or comma that opens a quoted phrase or a thesis, such as the quote before 'the arc AB' or the colon after 'Thesis statement'. Turning it up gave loose, repetitive text about the weather and the news.

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

    formed by the arc at the center of the circle. It is indicated by the⏎small letter M in front of the name. For example⏎is read as "the arc AB has a measure of 35 degrees".⏎See Arc Angle Measure

  2. #2 peak

    bad...As kill a king, and marry with his brother." Gertrude's immediate response is, "As kill a king?" What she means is, "What

  3. #3 peak

    I would like to write a paper about how headphones change the way people turn public space into their own private space. My thesis statement is ” How headphones demarcate personal space”. Headphones help demarcate personal space

  4. #4 peak

    out.⏎- Don’t engage with your thoughts. As Mooji says, “Let every thought come and hug you, but you don’t hug anything.”

  5. #5 peak

    structure is applicable for a general argumentative essay on abortion pro-choice.⏎- Define what abortion means⏎- Analyse the societal aspect of abortion⏎- Thesis statement: Should society’s contempt

  6. #6 peak

    early?► see thesaurus at seeNotice is not used in the progressive. You say: I notice you’re wearing glasses. ✗Don’t say: I’m noticing you’re wearing glasses.2 → be noticed/get (somebody

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: if this is a rule that I have to take that: I think that it's important that: - $ \boxed{ \, \, \, \, \, \, \, \, \, \, \, \

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

    in more than one point, then we see from Figure 2 that there are numbers and such that . This means that is not one-to-one. Therefore we have the following geometric method for determining whether a function is one-to-one. HORIZONTAL

  2. #2 peak

    impact on your eyes and thus it can be considered harmful to your health.⏎Why is blue light bad for the eyes? All light colors have a particular wavelength. And blue color is of the shortest wavelength. Due to its short wavelength and high energy

  3. #3 peak

    adjacent nodes (not including the node itself). Our depiction of graphs will look much like our depiction of trees. This should come as no surprise because, if you recall, trees are a special case of graphs.⏎If you cross-reference this chapter against

  4. #4 peak

    multi-lingual and interactive project - this means we need your input to ensure the pages are kept up to date. Find out how to contribute below.⏎Please note that this project is still being tested, if you have comments or suggestions, please

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. L13:895 suppresses
  2. L10:251 suppresses
  3. L11:1774 excites
  4. L9:1150 excites
  5. L6:2313 excites
  6. L11:2020 suppresses
  7. L15:2073 excites
  8. L14:2774 excites
  9. L7:1171 weakly excites
  10. L9:2247 weakly suppresses
  11. L12:1781 weakly suppresses
  12. L12:1380 weakly excites
  13. L14:1371 weakly suppresses unclear: document ends and instruction verbs
  14. L14:1902 weakly suppresses
  15. L15:2775 weakly excites
  16. L13:1521 weakly suppresses
  17. L12:247 weakly excites
  18. L14:531 weakly suppresses
  19. L9:2604 weakly excites
  20. L13:2672 weakly suppresses
  21. L15:1638 weakly suppresses
  22. L12:758 weakly excites
  23. L15:534 weakly suppresses
  24. L7:2369 weakly excites comma or 'to' before a quotation

Feeds later layers

  1. L18:1879 suppresses
  2. L18:2056 suppresses
  3. L18:2793 suppresses
  4. L19:1714 weakly suppresses
  5. L21:2565 weakly excites
  6. L24:373 weakly excites
  7. L17:2079 weakly excites
  8. L18:1971 weakly suppresses
  9. L17:2576 weakly suppresses
  10. L20:1300 weakly excites
  11. L20:2389 weakly excites
  12. L19:248 weakly suppresses
  13. L18:119 weakly suppresses
  14. L20:1614 weakly excites
  15. L24:711 weakly suppresses
  16. L19:2450 weakly excites
  17. L20:1539 weakly suppresses
  18. L18:2681 weakly excites
  19. L18:254 weakly suppresses
  20. L17:1463 weakly excites
  21. L17:1955 weakly suppresses
  22. L21:2800 weakly suppresses
  23. L21:1057 weakly excites
  24. L18:2657 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.

1 values between -2.98 and -2.851 values between -2.07 and -1.941 values between -1.94 and -1.811 values between -1.81 and -1.681 values between -1.42 and -1.291 values between -1.29 and -1.162 values between -1.16 and -1.035 values between -1.03 and -0.902 values between -0.90 and -0.7711 values between -0.77 and -0.6514 values between -0.65 and -0.5227 values between -0.52 and -0.39152 values between -0.39 and -0.26573 values between -0.26 and -0.131466 values between -0.13 and 0.001015 values between 0.00 and 0.13299 values between 0.13 and 0.26120 values between 0.26 and 0.3971 values between 0.39 and 0.5243 values between 0.52 and 0.6534 values between 0.65 and 0.7823 values between 0.78 and 0.9115 values between 0.91 and 1.048 values between 1.04 and 1.177 values between 1.17 and 1.305 values between 1.30 and 1.431 values between 1.43 and 1.562 values between 1.56 and 1.692 values between 1.69 and 1.822 values between 1.82 and 1.941 values between 1.94 and 2.072 values between 2.07 and 2.20offnegativepositiveoften

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. edException
  2. akespeare
  3. azines
  4. zsche
  5. alse
  6. ？」
  7. ignKey
  8. ество
  9. 其它问题
  10. ographed

Suppresses

  1. saying
  2. means
  3. ⏎⏎
  4. fact
  5. modo
  6. mem
  7. mean
  8. round
  9. :⏎⏎
  10. ⏎