Neuron Explorer

Qwen3-0.6B

unclear: deity names strongly negative

A neuron in layer 9 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

The positive side of this neuron is weak and scattered, with small hits on words like 'must have' and 'without'. Its strongest response is negative, on names of gods and deities such as 'gods', 'Lord' and 'Baal', but the steering results do not show a clear pattern.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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 small that are difficult to deal with". [Link]He must have been dumped by a small woman at some point in his life.<|endoftext|>"One death

  2. #2 peak

    beau, George Hayes, age unspecified, were sitting on the sill of an open second-floor window at 1 a.m., courting and sparking.⏎They must have sparked a little too energetically, because they both tumbled out the window

  3. #3 peak

    of x-rays was a happy accident. A physics professor named Wilhelm Roentgen was testing whether cathode rays could pass through glass and observed a mysterious light that would pass through human tissue and to reveal details of the bones and tissue underneath. Because

  4. #4 peak

    materials in this archive are copyright © The Manitoba Library Association. Reproduction for personal use is permitted only if this copyright notice is maintained. Any other reproduction is prohibited without permission Copyright information for reviewers⏎Young Canada Works<|endoftext|>From the Publisher:⏎Over roughly

  5. #5 peak

    with your counselor.⏎- Do ONE of the following:⏎- From a catalog of your choice, fill out an order form for merchandise as if you⏎intended to place an order. Share the completed form with your counselor and discuss⏎- With

  6. #6 peak

    . My trembling fingers clutched the cannister of bear spray fastened to my beltóa last resort in case of a grizzly charge. The bears were obviously aware of my presence, but upwind, and in the dark, they were

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 Greatness of the Lord" and the following: "The Lord of the Lord, the God of the Lord, who is the One of the God, the Lord who I will worship with all my life," and the following:

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

    Israel, causing hard feelings with his northern neighbor. He nevertheless won a major victory over the Edomites. Afterward, however, his veneration of the gods of Edom earned him a prophetic condemnation.⏎Seeking to punish Israel for

  2. #2 peak

    me to teach you to observe in the land that you are crossing the Jordan to possess, so that you, your children and their children after them may fear the Lord your God as long as you live by keeping all his decrees and commands that

  3. #3 peak

    house of Judah in the process, including Judah's King Ahaziah. Jehu also slaughtered the northern priests of Baal and put and end to Baal worship in his kingdom. However, this was

  4. #4 peak

    , on March 1, and succeeding days, carrying the Ancilia, they sang songs and hymns, and afterwards retired to a great banquet in the Temple of Mars. That the practice was originally Etruscan

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. L3:1978 excites
  2. L7:1269 excites
  3. L5:28 weakly excites
  4. L8:2825 weakly excites
  5. L2:2925 weakly excites vibr- word pieces (vibration)
  6. L4:2834 weakly excites
  7. L7:2207 weakly excites
  8. L6:1722 weakly excites
  9. L7:2945 weakly excites
  10. L6:873 weakly excites
  11. L2:319 weakly suppresses 'of' after 'pattern' or 'patterns'
  12. L8:1658 weakly suppresses
  13. L8:1082 weakly suppresses
  14. L6:2977 weakly suppresses
  15. L7:2430 weakly excites
  16. L8:1500 weakly excites
  17. L8:1704 weakly suppresses
  18. L3:1714 weakly suppresses
  19. L6:1702 weakly suppresses "not" or "instead" contrasting a correction
  20. L8:2097 weakly excites
  21. L5:2253 weakly suppresses
  22. L5:1583 weakly excites
  23. L0:3059 weakly excites
  24. L5:3034 weakly suppresses

Feeds later layers

  1. L18:2833 weakly suppresses
  2. L15:1672 weakly excites
  3. L23:1393 weakly suppresses
  4. L17:339 weakly suppresses
  5. L21:1008 weakly suppresses
  6. L26:994 weakly suppresses
  7. L11:2933 weakly suppresses
  8. L14:1685 weakly excites
  9. L11:751 weakly excites numbers and quantities mid-phrase
  10. L27:820 weakly suppresses
  11. L10:1801 weakly excites
  12. L12:121 weakly suppresses
  13. L21:2389 weakly suppresses
  14. L16:1770 weakly excites
  15. L22:2810 weakly suppresses
  16. L22:1468 weakly excites
  17. L15:201 weakly excites
  18. L26:2610 weakly excites
  19. L19:1913 weakly suppresses
  20. L22:1507 weakly suppresses
  21. L25:1319 weakly excites
  22. L23:2513 weakly excites
  23. L10:2155 weakly excites
  24. L20:2298 weakly 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.

1 values between -10.21 and -9.911 values between -6.65 and -6.361 values between -5.76 and -5.471 values between -3.39 and -3.092 values between -3.09 and -2.801 values between -2.50 and -2.211 values between -2.21 and -1.913 values between -1.91 and -1.611 values between -1.32 and -1.026 values between -1.02 and -0.7210 values between -0.72 and -0.43322 values between -0.43 and -0.132966 values between -0.13 and 0.17574 values between 0.17 and 0.4615 values between 0.46 and 0.762 values between 0.76 and 1.061 values between 1.35 and 1.65offnegativepositiveoften

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. xls
  2. PAL
  3. xls
  4. ineligible
  5. excel
  6. fy
  7. ,row
  8. endregion
  9. Denn
  10. ニア

Suppresses

  1. 祖先
  2. 帝
  3. wią
  4. лон
  5. 上帝
  6. christ
  7. крут
  8. Name
  9. name
  10. Zion