Neuron Explorer

Qwen3-0.6B

comma after an adjective in a list

A neuron in layer 9 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 comma or 'and' that comes right after a descriptive word, when a second description is about to follow, like 'tiny, round' or 'new, vibrant'. Turning it up made the model write long, repetitive, rambling text about the same idea.

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

    , teams of bold adventurers and astronomers set out for farflung corners of the planet to witness one of the rarest of all predictable astronomical events: The tiny, round silhouette of the planet Venus inching across the face of the sun.⏎Called

  2. #2 peak

    , as well as covering the role of the Irish language in Northern Ireland's more recent past. But the main body of the book is based on Malcolm's extensive and detailed research into the attitudes of young Protestants towards the Irish language, carried out

  3. #3 peak

    high amounts of fluorine. All 29 products that were tested for PFAS after showing high amounts of fluorine turned out to contain at least four specific, highly concerning PFAS. Some of these “forever chemicals” can break down

  4. #4 peak

    them throughout the period when other, similar regimes curtailed Jewish rights to curry favor with the Nazis. Still,

  5. #5 peak

    airi, a College of Health doctorate-candidate student, said that a survey found a 7 percent decrease in reports of students being bullied after the new, eight-week anti-bullying program.⏎Miyairi said that now is

  6. #6 peak

    Earth's Natural Rhythms⏎Yesterday's magnitude-7.2 earthquake that rocked northern Mexico and parts of Southern California was barely worth noting compared to the ongoing, major forces that shape planet Earth. In a geologic sense, there are greater

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 have been thinking about... [the time of day of day of the week... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...

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

    <|endoftext|>Microcontroller vs FPGA: Understanding the Key Differences⏎When it comes to embedded systems design, two technologies that often come up are microcontrollers and field-programmable gate arrays (FPGAs). While both serve similar purposes, they differ

  2. #2 peak

    roofs of houses. Polycrystalline silicon was by now the most common material for solar panels. There was one problem. Polycrystalline silicon had to be pure. This made them expensive. During the second decade of

  3. #3 peak

    complex problems, including poverty, as well as its ability to generate high-quality economic development.⏎The study also notes that higher poverty rates translate into higher costs for taxpayer-funded public assistance programs.⏎- Download the study, “Wal-Mart and County-Wide

  4. #4 peak

    be configured to work with a variety of feeding systems, including continuous web.⏎How does Plasma Compare to Corona?Corona treatment is an electrical process that uses ionized air to increase the surface tension of non-porous substrates. Normally, cor

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. L7:400 strongly excites
  2. L8:386 excites
  3. L8:1847 excites
  4. L7:2308 excites comma inside stacked adjectives
  5. L5:1851 suppresses
  6. L7:736 excites
  7. L8:526 suppresses
  8. L8:2099 excites
  9. L8:298 excites
  10. L8:725 suppresses
  11. L6:479 excites
  12. L4:1587 excites
  13. L7:1313 excites
  14. L2:1126 suppresses
  15. L5:1372 excites
  16. L7:2164 suppresses period inside country abbreviations
  17. L4:1156 excites
  18. L7:616 excites
  19. L5:2597 suppresses
  20. L7:2796 suppresses
  21. L5:595 excites
  22. L5:615 excites
  23. L7:1996 suppresses
  24. L8:1849 suppresses

Feeds later layers

  1. L13:976 excites
  2. L10:991 excites
  3. L10:578 excites
  4. L18:884 excites
  5. L10:959 suppresses
  6. L10:1940 weakly excites
  7. L11:1278 weakly excites
  8. L13:975 weakly excites
  9. L19:1657 weakly excites
  10. L12:999 weakly excites
  11. L12:431 weakly excites
  12. L13:308 weakly suppresses
  13. L13:1865 weakly excites
  14. L17:1280 weakly excites
  15. L19:2397 weakly suppresses
  16. L18:538 weakly excites
  17. L19:1412 weakly excites
  18. L15:499 weakly excites
  19. L22:837 weakly suppresses
  20. L10:67 weakly excites
  21. L15:702 weakly suppresses
  22. L11:233 weakly excites
  23. L24:1615 weakly excites
  24. L18:1228 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.

3 values between -1.87 and -1.671 values between -1.67 and -1.471 values between -1.47 and -1.274 values between -1.27 and -1.073 values between -1.07 and -0.874 values between -0.87 and -0.6735 values between -0.67 and -0.47329 values between -0.47 and -0.271235 values between -0.27 and -0.071799 values between -0.07 and 0.13271 values between 0.13 and 0.3384 values between 0.33 and 0.5338 values between 0.53 and 0.7323 values between 0.73 and 0.9320 values between 0.93 and 1.1313 values between 1.13 and 1.336 values between 1.33 and 1.538 values between 1.53 and 1.7410 values between 1.74 and 1.944 values between 1.94 and 2.144 values between 2.14 and 2.343 values between 2.34 and 2.543 values between 2.54 and 2.741 values between 2.74 and 2.941 values between 2.94 and 3.142 values between 3.34 and 3.541 values between 3.94 and 4.141 values between 5.54 and 5.741 values between 5.94 and 6.15offnegativepositiveoften

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. ulist
  2. sized
  3. phalt
  4. uez
  5. ward
  6. andidate
  7. .Slf
  8. Sanford
  9. ordable
  10. aybe

Suppresses

  1. …
  2. 其他人
  3. bris
  4. Did
  5. 另一位
  6. ...
  7. …"⏎⏎
  8. ..."⏎⏎
  9. isi
  10. ..."