Neuron Explorer

Qwen3-0.6B

the modal word 'may'

A neuron in layer 25 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 'may' when it means 'might be possible', as in 'there may be' or 'may occur'. When it was turned down, the text got stuck repeating 'may' or 'likely', and when it was turned up the text drifted into words like 'pain'.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    syysk. 2021|⏎|OKM-julkaisutyyppi||A1 Julkaistu artikkeli, soviteltu|<|endoftext|>Though it straddles the Arizona and Utah borders, Monument

  2. #2 peak

    regimes have short residence times, as Presidents come and go and control of Congress swings back and forth between parties. But ambitious astronomy—the kind that rewrites textbooks—may take decades to plan and execute. Building a

  3. #3 peak

    web site to request a printed copy of this publication.⏎- Our on-line publications are scanned and captured using Adobe Acrobat. During the capture process some typographical errors may occur. Please contact Sharon Hobrla, email@example.com if you notice any

  4. #4 peak

    thought that beta cells are completely wiped out (though some early research suggests there may still be some faint activity of beta cells in some people with type 1).⏎

  5. #5 peak

    are often not very precise. A more interesting question is what is the optimal type of motion.⏎I'm with you Naperlou. I would think there may be better vision models in nature than the human eye movement. The insect or bird

  6. #6 peak

    they have left their normal gathering grounds in⏎the Sea of Cortez in Baja California, Mexico. Fishermen worry that⏎a critical part of their livelihood may be gone. Many families have⏎depended on Humboldt squid since the slippery

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: may positional positional positional positional positional positional positional positional positional positional positional positional positional positional positional positional positional positional positional positional positional positional positional pos

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

    12, 2013⏎Nominated early this year for recognition on the UNESCO World Heritage List, which includes such famous cultural sites as the Taj Mahal, Machu Picchu and Stonehenge, the earthen works

  2. #2 peak

    ει τετράγωνον σχῆμα τῷ λόφῳ in Einklang zu bringen, ist es unumgänglich nothwendig dass der Ausdruck Roma quadrata

  3. #3 peak

    ” earthquakes, studies on active fault zones in Thailand are still minimal.⏎With all of these in mind, the Geophysics research group at Faculty of Science, Mahidol University is founded with the objective of being the “leader” in ge

  4. #4 peak

    incorrectly, an oil-analysis program is a waste of time and money. Key guidelines include:⏎- Take warmed-up machine samples while equipment is running. If that’s not possible, sample no later than 30 minutes after shutdown⏎- Use clean

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. L23:1449 strongly suppresses 'as' in 'as ... as possible'
  2. L21:2006 excites "as ... as possible" phrasing
  3. L24:766 excites
  4. L21:2227 excites
  5. L22:1486 excites
  6. L22:3060 excites
  7. L18:2565 excites
  8. L20:2868 suppresses
  9. L19:2765 excites
  10. L23:1342 suppresses
  11. L17:1765 weakly excites
  12. L15:721 weakly suppresses
  13. L22:2775 weakly excites
  14. L20:2994 weakly excites
  15. L23:470 weakly suppresses
  16. L4:693 weakly excites
  17. L17:1279 weakly excites
  18. L2:180 weakly excites 'is' after focus or goal nouns
  19. L22:2524 weakly excites
  20. L20:2810 weakly suppresses
  21. L21:1827 weakly suppresses
  22. L22:1104 weakly suppresses
  23. L14:280 weakly excites
  24. L18:2247 weakly suppresses

Feeds later layers

  1. L26:2591 suppresses
  2. L27:1630 suppresses
  3. L27:2029 suppresses
  4. L27:955 suppresses
  5. L27:1780 weakly suppresses
  6. L27:2303 weakly suppresses
  7. L26:1677 weakly suppresses
  8. L27:774 weakly excites
  9. L27:993 weakly excites
  10. L27:1699 weakly excites
  11. L27:2642 weakly suppresses
  12. L26:1960 weakly suppresses
  13. L26:2188 weakly suppresses
  14. L27:1479 weakly excites
  15. L27:2985 weakly excites
  16. L26:1855 weakly excites
  17. L26:466 weakly excites
  18. L26:3005 weakly excites
  19. L27:2602 weakly excites
  20. L26:2744 weakly excites commas inside lists before 'etc'
  21. L26:1304 weakly suppresses
  22. L26:2325 weakly excites
  23. L27:873 weakly excites
  24. L26:881 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 -11.20 and -10.611 values between -8.23 and -7.631 values between -6.44 and -5.851 values between -5.85 and -5.252 values between -5.25 and -4.661 values between -4.66 and -4.061 values between -4.06 and -3.474 values between -3.47 and -2.873 values between -2.87 and -2.285 values between -2.28 and -1.6823 values between -1.68 and -1.09394 values between -1.09 and -0.492709 values between -0.49 and 0.10692 values between 0.10 and 0.7040 values between 0.70 and 1.2910 values between 1.29 and 1.899 values between 1.89 and 2.493 values between 2.49 and 3.081 values between 3.08 and 3.682 values between 4.27 and 4.872 values between 4.87 and 5.461 values between 10.22 and 10.822 values between 12.01 and 12.60offnegativepositiveoften

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. 怎么说
  2. ailable
  3. urable
  4. 습
  5. смотреть
  6. 乎
  7. ません
  8. 贯通
  9. いう
  10. 纪委监

Suppresses

  1. 可能
  2. might
  3. may
  4. 可能会
  5. pos
  6. might
  7. likely
  8. 可能是
  9. 可能存在
  10. May