Neuron Explorer

Qwen3-0.6B

unclear: spots where text was removed

A neuron in layer 10 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

The strongest examples are places where a word or phrase seems to have been blanked out or marked as missing, such as an empty bracket right before a gap. The pattern is hard to pin down. Turning it up led to repetitive text that did not clearly relate to this.

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

    the air space with a soft polymer. The result was a continuous reinforced carbon nanotube composite.⏎“People have been working for more than ten years intensively [on carbon nanotube composites],” Suhr said. “Nobody had

  2. #2 peak

    far more effective at killing all cancerous cells than removing the entire breast.⏎Lead researcher Dr E Shelley Hwang, of the Duke Cancer Institute in <a class=”StrictlyAutoTagAnchor” title=”View all articles about North C<|endoftext|>What

  3. #3 peak

    in a school they believe did not support him.⏎Miyairi created the new anti-bullying program specifically to address these issues.⏎"It's important for [students] to understand the media messages and how our society views men and women,"

  4. #4 peak

    360 photos, edit them, design them and enhance the 360 space,” Ramnarine said. “Then [we] put those [photos] into virtual reality so you can experience the space after editing them.”⏎Other

  5. #5 peak

    1-day siege that included several gun battles with federal officers. Means died in 2012 at age 72.⏎“He was the only person in (a) leadership position in Wounded Knee who never left Wounded Knee,

  6. #6 peak

    behavior is easy. It’s maintaining change that’s hard—creating new and sustained ways of thinking and behaving. As Mark Twain quipped, “Quitting [smoking] is easy, I’ve done it dozens of times.” Many can

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: I am trying to write something about the things I want to do next time I feel something similar to a feeling of something specific. I want to write something that is specific and vivid' and I want to make it vivid' and specific'

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

    32K).⏎GLib (pronounced ``gee-lib'') shared libraries are a way to implement shared libraries on the Pilot that differs from the ``standard'' (SysLib) mechanism for the Palm Pilot.⏎Some features of GLib

  2. #2 peak

    , and the University of Colorado. UC Davis’s press release explains that the chemical’s effects are so striking that the study “provides strong evidence that triclosan could have effects on animal and human health at

  3. #3 peak

    free-market fundamentalism has become not only a much-vaunted ideology that now shapes all aspects of life in the United States but also a predatory global phenomenon “that drives the practices and principles of the International Monetary Fund, the World Bank, and

  4. #4 peak

    Atmospheric Administration have concluded that Louisiana is in line for the highest rate of sea-level rise “on the planet.” * Indeed, the water is rising so fast that some coastal restoration

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. L3:1388 strongly excites
  2. L9:632 strongly suppresses
  3. L6:174 excites
  4. L6:120 excites
  5. L9:982 suppresses
  6. L3:10 excites
  7. L7:100 suppresses
  8. L7:574 suppresses
  9. L7:732 excites
  10. L3:33 suppresses
  11. L2:240 suppresses
  12. L9:2249 suppresses
  13. L4:1861 suppresses
  14. L4:1005 suppresses
  15. L9:1376 weakly suppresses
  16. L4:2293 weakly excites
  17. L8:1380 weakly excites
  18. L4:327 weakly suppresses
  19. L1:102 weakly excites
  20. L6:80 weakly excites
  21. L0:649 weakly excites
  22. L1:122 weakly suppresses 'World War II' and the 'II' after 'War'
  23. L6:599 weakly suppresses
  24. L1:155 weakly suppresses

Feeds later layers

  1. L15:650 strongly suppresses
  2. L19:471 strongly suppresses
  3. L17:371 strongly suppresses
  4. L16:332 strongly suppresses plural '(s)' in parentheses
  5. L17:238 strongly suppresses
  6. L13:3050 suppresses
  7. L18:1705 suppresses
  8. L11:1579 excites
  9. L24:114 suppresses
  10. L19:671 suppresses
  11. L18:678 suppresses
  12. L19:588 suppresses
  13. L23:288 suppresses
  14. L20:540 suppresses
  15. L14:220 suppresses
  16. L15:2795 suppresses
  17. L14:428 suppresses
  18. L20:654 suppresses
  19. L21:210 excites
  20. L24:785 suppresses
  21. L12:2021 suppresses
  22. L25:372 suppresses
  23. L23:260 suppresses
  24. L14:1142 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 -5.79 and -5.621 values between -3.34 and -3.162 values between -2.81 and -2.646 values between -2.64 and -2.462 values between -2.46 and -2.292 values between -2.29 and -2.114 values between -2.11 and -1.947 values between -1.94 and -1.764 values between -1.76 and -1.599 values between -1.59 and -1.4111 values between -1.41 and -1.2414 values between -1.24 and -1.0622 values between -1.06 and -0.8948 values between -0.89 and -0.7188 values between -0.71 and -0.53132 values between -0.53 and -0.36241 values between -0.36 and -0.18515 values between -0.18 and -0.011044 values between -0.01 and 0.171090 values between 0.17 and 0.34585 values between 0.34 and 0.5274 values between 0.52 and 0.693 values between 0.69 and 0.873 values between 1.04 and 1.22offnegativepositiveoften

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. ⏎⏎⏎⏎⏎⏎⏎⏎⏎
  3. 分明
  4. `_
  5. “
  6. ⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎
  7. 、“
  8. ⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎⏎
  9. ⏎⏎⏎⏎⏎⏎⏎⏎
  10. (`

Suppresses

  1. 」と
  2. 》
  3. "},
  4. "}
  5. ”的
  6. 」
  7. ",
  8. ”。⏎⏎
  9. '",
  10. "]