Neuron Explorer

Qwen3-0.6B

'private' as in private property or schools

A neuron in layer 6 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 'private', especially in news-style phrases about private property, private schools or private hospitals. Its negative side fires on the word 'space' in the sense of a keyboard space bar or a gap between characters. Turning it up did not produce anything about private things.

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

    Carolina, Virginia, and throughout the South for refusing to honor the racially discriminatory exercise of private property rights by the owners of lunch counters.⏎What had been, as one contemporary

  2. #2 peak

    to reduce and reallocate public school funding to private schools. 70% of Americans oppose private school vouchers, and CEC opposes too!⏎Public schools continue to face the challenges

  3. #3 peak

    . Led by Dr. Tim Brenneman, department of plant pathology at the UGA Tifton campus, the program offers “any breeder, public or private, the opportunity to send germplasm they wish to have evaluated. The information

  4. #4 peak

    .⏎Health care workers from private hospitals, who are willing to get inoculated, can also get themselves administered with

  5. #5 peak

    fully accountable to the same accountability measures as traditional district schools. IDEA will receive funding from a number of federal, state, and local government agencies as well as from private and public foundations and individual philanthropists whose investments illustrate their support for IDEA’s

  6. #6 peak

    schooling, findings showed a push for more public charter schools as well as allowing parents to home school their children. However, there was strong opposition when it came to private school vouchers which would put public education in direct competition with private education, threatening

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 term "bachelor" is a common term in the context of the English language. What does the term "bachelor" mean in the context of a person's life? A. The term "bachelor" is a common term

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

    how this works⏎Tip: insert a question mark by pressing your spacebar.⏎Definition of: ALLY⏎- (n.) See Alley, a

  2. #2 peak

    example.⏎In (most of) the UK, geographic locations are commonly identified by a postcode. This is a two-part set of letters and digits separated by a space, designed in the 1950s to assist

  3. #3 peak

    is a \" quote."; would print This is a " quote without escaping the quote in the middle you would receive an error.⏎When a filename or directory needs a space in Linux you can use a backward slash followed by a space, to escape the

  4. #4 peak

    should optionally report changes to special files specially,⏎patch should be extended to understand these extensions.⏎When a file name contains an unusual character like a newline or⏎white space, `diff -r' generates a patch that⏎parse. The problem

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. L4:25 weakly excites
  2. L4:1243 weakly suppresses
  3. L5:405 weakly suppresses
  4. L1:1595 weakly suppresses
  5. L4:179 weakly suppresses
  6. L4:245 weakly excites
  7. L4:19 weakly suppresses
  8. L5:356 weakly excites
  9. L4:182 weakly suppresses
  10. L0:1793 weakly excites
  11. L4:21 weakly excites
  12. L5:102 weakly excites
  13. L4:88 weakly suppresses
  14. L1:2738 weakly excites
  15. L4:37 weakly suppresses
  16. L5:1815 weakly excites
  17. L4:71 weakly suppresses
  18. L0:642 weakly excites
  19. L0:2679 weakly excites
  20. L5:809 weakly excites
  21. L3:2576 weakly suppresses
  22. L5:28 weakly excites
  23. L3:1310 weakly excites
  24. L3:2310 weakly suppresses

Feeds later layers

  1. L7:1843 suppresses
  2. L7:1642 weakly suppresses
  3. L27:1756 weakly excites
  4. L7:526 weakly suppresses
  5. L11:2212 weakly excites
  6. L14:771 weakly suppresses
  7. L8:631 weakly excites
  8. L9:306 weakly excites
  9. L9:1955 weakly excites
  10. L7:2233 weakly excites
  11. L20:841 weakly excites
  12. L20:101 weakly excites
  13. L27:3047 weakly suppresses
  14. L26:2532 weakly excites
  15. L14:492 weakly excites
  16. L12:1785 weakly excites
  17. L22:1326 weakly excites
  18. L25:670 weakly suppresses
  19. L12:1595 weakly excites
  20. L8:1487 weakly excites
  21. L11:2687 weakly excites
  22. L22:1740 weakly suppresses
  23. L16:2397 weakly excites
  24. L7:2374 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.

1 values between -1.53 and -1.441 values between -0.98 and -0.897 values between -0.52 and -0.4219 values between -0.42 and -0.3374 values between -0.33 and -0.24281 values between -0.24 and -0.15894 values between -0.15 and -0.051553 values between -0.05 and 0.04725 values between 0.04 and 0.13244 values between 0.13 and 0.2276 values between 0.22 and 0.3222 values between 0.32 and 0.415 values between 0.41 and 0.504 values between 0.50 and 0.591 values between 0.59 and 0.691 values between 2.07 and 2.16offnegativepositiveoften

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. ӣ
  2. scars
  3. getChild
  4. ostringstream
  5. Infinity
  6. connections
  7. \Modules
  8. 披
  9. met
  10. 的设计

Suppresses

  1. 역
  2. icult
  3. ulls
  4. לבן
  5. imped
  6. ixe
  7. 榈
  8. yg
  9. rif
  10. urities