Neuron Explorer

Qwen3-0.6B

the word 'can' after a subject

A neuron in layer 17 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 'can' right after a noun or pronoun subject, as in 'Hives can occur' or 'It can be challenging'. Turning it up did not make the output obviously about ability or possibility.

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

    which is an examination of what motivates individuals. According to Maslow, people must achieve basic survival needs before reaching one’s full potential.⏎Maslow’s theory can easily be applied to workplace mentality that runs the gamut of employees just showing up

  2. #2 peak

    is experiencing a hypersensitivity reaction. Fluid escaping from blood vessels becomes trapped in parts of the skin and lining membranes of the body, causing localized swelling.⏎Hives can occur anywhere on the body where there is skin. Therefore, you can also get

  3. #3 peak

    degree the potential for addiction. In this article I want to explain what addiction is and how gambling fits in.⏎It can lead to the establishment of a powerful and difficult to break state<|endoftext|>Even before the

  4. #4 peak

    people consider “food allergy” may instead be a food intolerance or non-allergic issue. Yet, food allergy remains an important problem in children and adults, and can be life threatening. Methods of testing (including skin testing and blood testing) for

  5. #5 peak

    bond, namely the coupon payment throughout the life of the bond and the principal payment, or the balloon payment, at the end of the bond's life. You can see how it changes over time in the bond price chart in our calculator.⏎To

  6. #6 peak

    , V, X, L, C, D, M.⏎In conclusion, we simply find the significant difference between the Arabic and Roman numerals. This difference can certainly be overcome with the numeral converter.⏎Why Arabic to Roman Numeral Converter?⏎

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: I am the 4th and 5th and 5th and 5th and 5th and 5th and 5th and 5th and 10th and 10th and 10

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

    ! Click on one of the images to start.<|endoftext|>Appendicitis is a serious medical condition that can be life-threatening if not treated promptly.While appendicitis is most common in people between the ages of 15 and 30,

  2. #2 peak

    , either permanently, as a hermit or monk; or partially, as every member of a religious institute; or temporarily,k as in a retreat. Physical solitude is not escapism or isolationism, but a means to an end, the end

  3. #3 peak

    PCs will make copies in the process, and the website host grants visitors an implied license or permission to make those copies.4⏎Downloading content from any web page is the equivalent to making a copy of the content, the same as making copies of

  4. #4 peak

    in the world. Physicians and researchers who specialize in the study, diagnosis, treatment, and prevention of cancer are called oncologists.⏎What causes cancer?⏎Cancer is ultimately the result of cells that uncontrollably grow and do not die. Normal

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. L10:384 strongly excites
  2. L4:173 strongly suppresses
  3. L3:978 suppresses
  4. L15:886 excites
  5. L10:2166 excites
  6. L0:1372 suppresses
  7. L13:2339 excites
  8. L11:360 suppresses
  9. L6:170 excites
  10. L5:387 excites
  11. L9:352 suppresses
  12. L6:1363 suppresses
  13. L4:904 excites
  14. L3:2961 suppresses
  15. L6:7 weakly suppresses
  16. L16:2349 weakly excites
  17. L5:1692 weakly suppresses
  18. L5:2772 weakly excites
  19. L5:371 weakly excites
  20. L16:968 weakly suppresses
  21. L4:693 weakly excites
  22. L13:1542 weakly suppresses
  23. L8:4 weakly excites
  24. L8:614 weakly excites

Feeds later layers

  1. L24:2814 strongly excites the modal 'can' after 'there' or 'it'
  2. L18:2896 strongly excites
  3. L18:1491 strongly excites
  4. L19:2701 strongly excites
  5. L19:1473 strongly excites "can" after a noun in a clause
  6. L21:2007 excites
  7. L19:2433 excites
  8. L20:2487 excites
  9. L18:522 excites
  10. L22:1892 excites the digit 6 in years 19xx
  11. L19:1709 suppresses
  12. L26:1452 excites
  13. L19:1850 suppresses
  14. L20:636 excites
  15. L19:2185 excites
  16. L18:2634 suppresses
  17. L19:1294 excites
  18. L18:736 excites
  19. L21:853 excites 'can' after 'and' or a noun subject
  20. L24:298 excites 'would' in hypothetical or past statements
  21. L18:2044 excites
  22. L18:262 excites
  23. L24:864 excites 'to' before an infinitive
  24. L22:1422 suppresses

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.

336 values between -2.53 and -1.143457 values between -1.14 and 0.2646 values between 0.26 and 1.6516 values between 1.65 and 3.0512 values between 3.05 and 4.447 values between 4.44 and 5.843 values between 5.84 and 7.236 values between 7.23 and 8.631 values between 10.02 and 11.422 values between 11.42 and 12.811 values between 12.81 and 14.203 values between 14.20 and 15.601 values between 19.78 and 21.182 values between 21.18 and 22.571 values between 25.36 and 26.761 values between 26.76 and 28.151 values between 28.15 and 29.541 values between 29.54 and 30.941 values between 30.94 and 32.331 values between 32.33 and 33.731 values between 33.73 and 35.121 values between 35.12 and 36.521 values between 37.91 and 39.311 values between 40.70 and 42.093 values between 42.09 and 43.491 values between 50.46 and 51.861 values between 51.86 and 53.25offnegativepositiveoften

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. 't
  2. ’t
  3. |R
  4. easily
  5. ’T
  6. afford
  7. safely
  8. |
  9. 很好地
  10. adians

Suppresses

  1. 网讯
  2. apsed
  3. 了一批
  4. esModule
  5. иж
  6. 示范基地
  7. 骥
  8. bases
  9. дум
  10. lest