Neuron Explorer

Qwen3-0.6B

capital F starting a word

A neuron in layer 27 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 responds to a capital F that begins a word or heading, like 'Fishing', 'Feminist', or 'Ferrets', especially where the word is split off from the text before it. Turning it up or down gave text with no clear link to that letter, so the effect on writing is hard to read.

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

    teachers to introduce children to the enchantment and magicality found in words. Highly recommended. 5 Stars. First reviewed for Readers Favorite.⏎by Fiona Ingram<|endoftext|>Fishing is both an art and a science. It is an art because it requires

  2. #2 peak

    life trajectories, goals, and challenges; attentiveness to social, institutional, and individual power structures and their relationships to privilege,⏎What is Feminist Family Therapy?⏎Feminist Family Therapy offers case-based discussions that address race, gender, and class

  3. #3 peak

    play an important role in the Israel-Palestine conflict, Freedland suggests examining the role Irish Americans had in helping to settle the conflict in Northern Ireland.⏎Freedland writes, the role of the Irish-American diaspora in the Northern Ireland

  4. #4 peak

    consort of Castile and León|⏎|Spouse||Alfonso X of Castile|⏎|Beatrice, Marchioness of Montferrat⏎Ferdinand de la Cerda⏎Sancho IV of Castile⏎|Father

  5. #5 peak

    asonic transmitters in French Polynesia between 1985 and 1997. Nine fish were caught near Fish Aggregating Devices (FADs) while the other two were tracked in coastal areas without FADs

  6. #6 peak

    ne, José Naranjo, Veterinary Doctor and international consultant in Animal Health and Public Health, provided an interesting analysis of the impact of foot-and-mouth disease (FMD) on countries’ production and

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 whole middle part wasn’t part of the whole whole part part part part part part part part part part part part part part part part part part part part part part part part part part part part part part part part part part part part part

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

    ptomatically.⏎-Please note this is not a medical mask and should not be worn as personal protective equipment.⏎-The face covering does not guarantee safety or protection.⏎-Face coverings should not be used as a replacement for social distancing or self isolation

  2. #2 peak

    40|⏎|Monday||May 01||Labour Day||Youm-e-Mazdoor|⏎|Monday||June 26||Eid-ul-Fitr||The end of Ramadan|⏎|Tuesday||June 27||E

  3. #3 peak

    The Value of Everything: Making and Taking in the Global Economy by Mariana Mazzucato. Copyright © 2018. Available from PublicAffairs, an imprint of Perseus Books, LLC, a subsidiary of H

  4. #4 peak

    January 8, by nadmin. Seminar for AAC – Cleveland This distrust is transferred to every person the adoptee wants to get close to. Self-blame/failure/low self-esteem Sadness/depression: loss . In Building Self-E

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. L22:369 excites
  2. L23:2441 excites
  3. L18:437 excites
  4. L20:211 excites
  5. L17:378 excites
  6. L0:755 suppresses
  7. L18:328 suppresses
  8. L2:1622 excites
  9. L26:1042 suppresses
  10. L7:300 excites
  11. L17:253 weakly suppresses
  12. L19:1147 weakly suppresses
  13. L22:961 weakly suppresses
  14. L21:2605 weakly excites
  15. L1:9 weakly excites
  16. L16:47 weakly excites
  17. L0:1365 weakly suppresses
  18. L9:449 weakly suppresses
  19. L16:67 weakly excites
  20. L17:543 weakly suppresses
  21. L4:1892 weakly excites
  22. L16:11 weakly suppresses
  23. L19:702 weakly excites
  24. L7:49 weakly suppresses

Feeds later layers

None: this is the last layer.

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 -18.81 and -18.021 values between -17.22 and -16.432 values between -14.04 and -13.251 values between -13.25 and -12.451 values between -11.66 and -10.861 values between -5.29 and -4.503 values between -4.50 and -3.701 values between -3.70 and -2.914 values between -2.91 and -2.114 values between -2.11 and -1.3271 values between -1.32 and -0.523529 values between -0.52 and 0.28255 values between 0.28 and 1.0714 values between 1.07 and 1.875 values between 1.87 and 2.664 values between 2.66 and 3.463 values between 3.46 and 4.254 values between 4.25 and 5.051 values between 5.05 and 5.841 values between 5.84 and 6.642 values between 12.21 and 13.00offnegativepositiveoften

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. ways
  2. choice
  3. choix
  4. spur
  5. cause
  6. 引起
  7. unders
  8. 见效
  9. cause
  10. 琵

Suppresses

  1. plits
  2. orce
  3. orea
  4. rather
  5. un
  6. powstał
  7. ptides
  8. eel
  9. Bloody
  10. corner