Neuron Explorer

Qwen3-0.6B

"of" after full text or version

A neuron in layer 8 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

Fires on the word "of" in phrases like "full text of", "version of" or "copy of", where the next words will name a document or edition. Turning it up made the model's writing wander off into unrelated topics, so the link to the steering is weak.

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

    As a whole, these examples illustrate the three kinds of discourse at which rhetoric is aimed.⏎Oxford Scholarship Online requires a subscription or purchase to access the full text of books within the service. Public users can however freely search the site and view the

  2. #2 peak

    retrotransposons.⏎Retrotransposons are small bits of genetic information that can move around and multiply themselves in the genome via RNA (a molecule similar to DNA). Imagine the DNA in your cells as a long sentence and retrotranspos

  3. #3 peak

    diagnosis of children presenting with polio-like symptoms.⏎- AAP News: 'AAP clinical report outlines how US pediatricians can help stop polio'⏎- Full text of Clinical Report⏎East African pediatricians speak for one for vaccines⏎

  4. #4 peak

    cannot be saved, field biologists help bring back species and their ecosystems (restoration biology).⏎Chicago Scholarship Online requires a subscription or purchase to access the full text of

  5. #5 peak

    Cooper.⏎Buy the hardcover edition of⏎I Heard the Willow Weep for $6.00. (Retail

  6. #6 peak

    eors from the on-board camera will be available online at the Space Sciences Website at: http://www.perseidslive.com⏎For an electronic version of this advisory or more information, visit Marshall’s News Center Website at:http://www

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: A user and a team lead to a new position in a company. The company has a policy of not allowing employees and contractors and internal auditors and contractors and managers and managers and managers and managers and managers and managers and managers and managers and

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

    Professor Atkinson said the latest wave of automation in the mining industry came in the form of

  2. #2 peak

    Gerald R. Ford.⏎Transcript from the February 10, 199

  3. #3 peak

    University Giessen.⏎Gabriel and Lucifer⏎This English translation of a poem by the poet

  4. #4 peak

    portal/species/overview?p_p_spp=257336⏎This

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. L3:437 suppresses
  2. L1:1371 weakly suppresses
  3. L6:104 weakly excites
  4. L3:875 weakly suppresses
  5. L5:261 weakly suppresses
  6. L1:148 weakly suppresses
  7. L4:305 weakly excites
  8. L2:1945 weakly suppresses
  9. L5:1927 weakly suppresses
  10. L3:510 weakly suppresses
  11. L4:220 weakly excites
  12. L3:1717 weakly suppresses
  13. L1:1919 weakly suppresses
  14. L1:388 weakly suppresses
  15. L3:684 weakly excites
  16. L1:138 weakly excites
  17. L4:237 weakly excites
  18. L3:293 weakly excites
  19. L3:1225 weakly excites
  20. L0:1494 weakly suppresses
  21. L1:386 weakly suppresses
  22. L4:535 weakly suppresses
  23. L3:2458 weakly excites
  24. L7:1443 weakly suppresses

Feeds later layers

  1. L10:55 suppresses
  2. L13:473 weakly excites
  3. L9:491 weakly suppresses
  4. L16:3034 weakly suppresses
  5. L17:666 weakly suppresses
  6. L27:1611 weakly suppresses
  7. L10:2 weakly suppresses
  8. L18:2233 weakly suppresses
  9. L14:738 weakly suppresses
  10. L9:630 weakly excites
  11. L9:1131 weakly suppresses
  12. L17:2225 weakly excites unclear: verbs after subject, line breaks
  13. L12:5 weakly suppresses
  14. L14:254 weakly suppresses
  15. L10:1398 weakly suppresses
  16. L9:1132 weakly suppresses
  17. L12:661 weakly excites
  18. L16:495 weakly suppresses
  19. L10:2376 weakly suppresses
  20. L16:1246 weakly suppresses
  21. L12:887 weakly excites
  22. L10:1797 weakly suppresses
  23. L16:324 weakly suppresses
  24. L24:81 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.

2 values between -3.57 and -3.451 values between -3.45 and -3.331 values between -3.33 and -3.211 values between -2.96 and -2.841 values between -2.60 and -2.471 values between -2.47 and -2.351 values between -2.23 and -2.111 values between -1.86 and -1.742 values between -1.74 and -1.622 values between -1.62 and -1.501 values between -1.37 and -1.251 values between -1.25 and -1.132 values between -1.13 and -1.015 values between -1.01 and -0.892 values between -0.89 and -0.764 values between -0.76 and -0.644 values between -0.64 and -0.5215 values between -0.52 and -0.4070 values between -0.40 and -0.28321 values between -0.28 and -0.15914 values between -0.15 and -0.031585 values between -0.03 and 0.09660 values between 0.09 and 0.21230 values between 0.21 and 0.3454 values between 0.34 and 0.4614 values between 0.46 and 0.587 values between 0.58 and 0.703 values between 0.70 and 0.821 values between 0.82 and 0.951 values between 0.95 and 1.071 values between 1.19 and 1.31offnegativepositiveoften

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. appearance
  2. روب
  3. diss
  4. onward
  5. disappeared
  6. current
  7. 圭
  8. 없
  9. desire
  10. .INSTANCE

Suppresses

  1. isted
  2. ising
  3. isable
  4. atement
  5. 注册资本
  6. ätz
  7. ocker
  8. illé
  9. raud
  10. gnore