Neuron Explorer

Qwen3-0.6B

'and' linking a role to a name

A neuron in layer 14 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 'and' when it joins a person's job title or role to their name, as in 'physicist and best-selling author Brian Greene'. Turning it up did not produce anything obviously related to that pattern.

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

    small schools in Michigan. HFLI developed a design thinking middle/high school curriculum. They have<|endoftext|>On Wednesday, protesters wrecked a statue of Philadelphia abolitionist and philanthropist Matthias Baldwin. They defaced the face of the statue, covered

  2. #2 peak

    urban oasis⏎Union Square began as part of Willowbrook, the John Donnell Federal-period estate, which he purchased in 1802 from Baltimore merchant and later Mayor Thorowgood Smith. In 1847, the Don

  3. #3 peak

    exhaust stream to generate a high energy steam cycle, which can then be used to help reduce the engine's fuel consumption.⏎Commenting on the projects, Project lead and Lecturer in the Department of Mechanical Engineering, Dr Copeland said: "These

  4. #4 peak

    .⏎Its analysis follows on from previous research by the think tank which found Greater Manchester has "lethal and illegal" levels of NO2 air pollution.⏎Report author and research fellow Jack Hunter said the well-being of people living in the North, and

  5. #5 peak

    increase calorie burn and reduce obesity risk in mice.⏎Apples Are Diverse⏎The apple is the most diverse food plant in the world. According to food journalist and apple expert Rowan Jacobsen, there used to be 16,0

  6. #6 peak

    create shade with a retractable awning, an umbrella or trees. "Look at the gifts and challenges that nature gives you in your own microclimate," architect and Natural Home columnist Carol Venolia says. "Working with nature is simple, but

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 I, but I, but I, but I, but I, but I, but I, but I, but I, but I, but I, but I, but I, but I, but I, but I

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

    of Christ scenes for distribution by Christian publishers.⏎Head of Christ. This famous Protestant icon began as a charcoal sketch composed in 1924 for the official magazine of Sallman's denomination. A publishing firm affiliated with the Church of God

  2. #2 peak

    widely in India. This happened during WW II. In 1942 he was active in an underground movement against colonial rule. He also edited the literary magazine `Tomorrow` for some days.⏎`Kanthapura` is the first

  3. #3 peak

    carbon fuel use⏎Ottawa, Canada, May 5, 2012: “Climate campaigners at 350.org and their partner organizations are completely off base with Climate Impacts Day,” said Tom

  4. #4 peak

    an official under the czar’s vicegerent in Warsaw. Rzewuski edited the government newspaper Dziennik warszawski.⏎WORKSDziela, vols.

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. L12:2947 strongly excites
  2. L13:1979 strongly excites
  3. L10:2935 strongly suppresses
  4. L6:1412 suppresses
  5. L12:2944 excites 'of' after a field, state or city noun
  6. L13:2403 excites
  7. L12:1662 suppresses
  8. L12:2989 suppresses
  9. L13:1420 suppresses
  10. L13:1140 excites
  11. L8:2006 excites
  12. L12:1670 excites
  13. L11:3062 suppresses
  14. L12:1811 suppresses
  15. L13:1384 excites
  16. L10:1627 excites
  17. L12:1393 excites
  18. L6:1639 excites
  19. L12:2599 excites
  20. L13:1376 excites
  21. L13:2311 suppresses
  22. L13:298 suppresses
  23. L13:838 suppresses
  24. L12:1184 excites

Feeds later layers

  1. L16:985 strongly suppresses
  2. L18:754 suppresses
  3. L19:761 suppresses
  4. L18:1678 suppresses
  5. L20:769 suppresses
  6. L15:1299 suppresses
  7. L20:331 excites
  8. L16:1599 suppresses
  9. L16:2067 suppresses
  10. L21:336 suppresses
  11. L17:255 suppresses
  12. L16:2178 suppresses
  13. L18:995 suppresses
  14. L17:775 suppresses
  15. L19:1934 suppresses
  16. L19:2576 suppresses
  17. L19:1017 suppresses
  18. L15:1497 suppresses
  19. L15:1381 suppresses
  20. L16:3012 suppresses
  21. L17:445 suppresses
  22. L18:92 suppresses
  23. L20:1112 suppresses
  24. L15:261 suppresses

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 -2.84 and -2.731 values between -2.73 and -2.621 values between -1.98 and -1.871 values between -1.66 and -1.551 values between -1.45 and -1.341 values between -1.34 and -1.233 values between -1.13 and -1.021 values between -0.91 and -0.808 values between -0.80 and -0.7017 values between -0.70 and -0.5950 values between -0.59 and -0.48158 values between -0.48 and -0.38291 values between -0.38 and -0.27531 values between -0.27 and -0.16733 values between -0.16 and -0.05731 values between -0.05 and 0.05470 values between 0.05 and 0.16313 values between 0.16 and 0.27242 values between 0.27 and 0.38168 values between 0.38 and 0.48107 values between 0.48 and 0.5944 values between 0.59 and 0.7019 values between 0.70 and 0.8010 values between 0.80 and 0.912 values between 0.91 and 1.023 values between 1.12 and 1.231 values between 1.34 and 1.45offnegativepositiveoften

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. strup
  2. WithValue
  3. FileVersion
  4. Doug
  5. !⏎⏎⏎⏎
  6. ług
  7. 被执行
  8. !⏎⏎⏎⏎⏎⏎
  9. -blocking
  10. LIKE

Suppresses

  1. азв
  2. oyer
  3. eller
  4. igma
  5. eters
  6. qrt
  7. X
  8. isé
  9. "
  10. clair