Neuron Explorer

Qwen3-0.6B

end of document, boilerplate footers

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 lights up at the end of a web page, right where the text stops and a new unrelated document begins, especially after copyright notices, contact lines or reference lists. When it was turned up, the model's text drifted into repetitive, tangled phrasing.

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

    ions in the region, and are insufficient to offset present rates of relative sea level rise. © 2006 Elsevier Ltd. All rights reserved.<|endoftext|>Are you tired of being sick and tired of being sick and tired? Then,

  2. #2 peak

    .⏎ASJC Scopus subject areas⏎- Electronic, Optical and Magnetic Materials⏎- Ceramics and Composites⏎- Condensed Matter Physics⏎- Materials Chemistry<|endoftext|>Humans are intimately connected with the physical environment, even though we have only been present

  3. #3 peak

    status||Published - 17 Oct 2008|⏎- age-related changes⏎- equid skulls⏎- horse use⏎- work-related changes<|endoftext|>Make your Arduino play the French Can Can using a buzzer. No pliers

  4. #4 peak

    created by Joanne Sutherland, a genetic counsellor at the Hospital for Sick Children's occular genetics program.⏎You can help – support us today!<|endoftext|>January 10th, 2004 01:1

  5. #5 peak

    LibGuide. Contact Sue Samson, firstname.lastname@example.org, with questions or for more information on how this database can help you with your research or teaching.<|endoftext|>Iris is technically the Greek goddess of messages, the sky, and the sea

  6. #6 peak

    our tutors. If you have any questions, please contact us at firstname.lastname@example.org or 410.837.5383.<|endoftext|>Microsoft co-founder Paul Allen has shown off the “

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: that the amount of variation in the family of functions that can be represented by this kind of linear motion (like this family of functions) as a constant factor, when it's embedded in the same level of order of magnitude (e.g.,

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

    Free Trial to Unlock the Full Review⏎Why Lesson Planet?⏎Find quality lesson planning resources, fast!⏎Share & remix collections to collaborate.⏎Organize your curriculum with collections. Easy!⏎Have time to be more creative & energetic with your students!⏎Students

  2. #2 peak

    Free Trial to Unlock the Full Review⏎Why Lesson Planet?⏎Find quality lesson planning resources, fast!⏎Share & remix collections to collaborate.⏎Organize your curriculum with collections. Easy!⏎Have time to be more creative & energetic with your students!⏎Students

  3. #3 peak

    location, we recommend that you select: .⏎You can also select a web site from the following list:⏎Select the China site (in Chinese or English) for best site performance. Other MathWorks country sites are not optimized for visits from your location

  4. #4 peak

    Code programming site, learn computer science using the CS Discoveries class on code.org, jump into CircuitPython to learn Python and hardware together, TinyGO, or even use the Arduino IDE. Circuit Playground Express is the newest and best Circuit

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. L10:80 excites
  2. L9:25 excites
  3. L8:32 weakly excites
  4. L9:257 weakly suppresses
  5. L15:975 weakly suppresses
  6. L16:2020 weakly excites
  7. L10:196 weakly excites
  8. L14:94 weakly excites
  9. L3:157 weakly excites
  10. L12:1166 weakly excites
  11. L1:1279 weakly suppresses
  12. L10:61 weakly suppresses
  13. L16:926 weakly excites
  14. L15:81 weakly suppresses line breaks inside running text
  15. L6:451 weakly excites line break before a subheading
  16. L13:53 weakly excites period ending a claim before elaboration
  17. L1:235 weakly suppresses
  18. L5:2084 weakly excites
  19. L16:611 weakly excites
  20. L16:655 weakly excites 'see' cross-references and web boilerplate
  21. L15:66 weakly excites
  22. L12:1442 weakly excites end of document after a full paragraph
  23. L14:758 weakly suppresses
  24. L3:177 weakly suppresses

Feeds later layers

  1. L18:442 excites
  2. L26:732 suppresses
  3. L27:59 excites full stops in running text
  4. L20:940 excites
  5. L19:230 suppresses
  6. L25:409 weakly suppresses
  7. L19:384 weakly excites
  8. L19:540 weakly excites
  9. L27:78 weakly excites
  10. L18:126 weakly suppresses comma joining a full clause
  11. L18:382 weakly excites
  12. L19:301 weakly excites
  13. L19:1984 weakly excites
  14. L19:308 weakly suppresses
  15. L19:1084 weakly suppresses "and" joining clauses after a condition
  16. L19:2085 weakly excites
  17. L23:1196 weakly excites
  18. L18:576 weakly excites
  19. L18:592 weakly suppresses
  20. L19:68 weakly excites full stops closing a claim
  21. L19:509 weakly excites
  22. L21:524 weakly excites
  23. L19:1455 weakly suppresses
  24. L23:1861 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.

2 values between -1.78 and -1.314 values between -1.31 and -0.8427 values between -0.84 and -0.382148 values between -0.38 and 0.091527 values between 0.09 and 0.5698 values between 0.56 and 1.0312 values between 1.03 and 1.4917 values between 1.49 and 1.965 values between 1.96 and 2.437 values between 2.43 and 2.904 values between 2.90 and 3.363 values between 3.36 and 3.831 values between 3.83 and 4.305 values between 4.30 and 4.773 values between 4.77 and 5.233 values between 5.23 and 5.704 values between 5.70 and 6.1713 values between 6.17 and 6.642 values between 6.64 and 7.105 values between 7.10 and 7.576 values between 7.57 and 8.041 values between 8.51 and 8.971 values between 9.91 and 10.381 values between 10.38 and 10.841 values between 10.84 and 11.311 values between 11.31 and 11.781 values between 11.78 and 12.251 values between 12.25 and 12.711 values between 13.18 and 13.652 values between 15.52 and 15.991 values between 15.99 and 16.451 values between 16.45 and 16.92offnegativepositiveoften

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. #####
  2. hp
  3. }}},⏎
  4. unn
  5. A
  6. entials
  7. umes
  8. ования
  9. iya
  10. "}}⏎

Suppresses

  1. .TryParse
  2. clearTimeout
  3. countryCode
  4. elsius
  5. asley
  6. StringEncoding
  7. EventType
  8. newIndex
  9. \Queue
  10. =-=-=-=-