Neuron Explorer

Qwen3-0.6B

'of' in 'School of' institution names

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 lights up on the word 'of' when it follows 'School' in names like 'School of Medicine' or 'School of Public Health'. It is specific to school names, and the steering results do not show any related theme.

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

    1921-School of Music organized.⏎1923-School of Business organized.⏎195

  2. #2 peak

    hn School of Medicine at Mount Sinai.<|endoftext|>Visualise different economic scenarios, how they might be

  3. #3 peak

    ayashi is Associate Professor in the Department of Media and Communication at City University of Hong Kong.⏎Atsushi Tago is Professor in the School of Political Science and Economics at Waseda University, Tokyo, Japan and Global Fellow

  4. #4 peak

    in an age of terrorism.⏎Kristen Tepfenhardt, The New York Times Learning Network⏎Yasmin Chin Eisenhauer, The Bank Street College of Education in

  5. #5 peak

    Harvard T.H. Chan School of Public Health, Sept. 2012, https://www.hsph

  6. #6 peak

    © 1994- Drexel University. All rights reserved.⏎The Math Forum is a research and educational enterprise of the Drexel University School of Education.<|endoftext|>Maybe you’ve tried intermittent fasting (IF) to shed a few

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 question is, What do you think, or maybe, if you are going to be a student, or someone who is considering, maybe, the importance of the following statement: "I. In the context of this, the importance of

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

    #1: tough ; ASCII code decremented by 32⏎String#2: HGUOT ; reverse order⏎mov char, al⏎L1: mov eax,32 ; creates ASCII code decrement by 32⏎mov

  2. #2 peak

    54.0.1 to 169.254.255

  3. #3 peak

    361. He moved his capital to Adrionople (now Edirne), Thr

  4. #4 peak

    67. He entered the medical course at the University of Tübingen, but the

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. L1:242 weakly excites
  2. L5:2250 weakly suppresses
  3. L7:3035 weakly suppresses
  4. L3:158 weakly suppresses
  5. L4:1221 weakly excites
  6. L4:22 weakly suppresses
  7. L8:144 weakly excites
  8. L8:18 weakly suppresses
  9. L7:866 weakly excites
  10. L4:181 weakly suppresses
  11. L0:50 weakly excites
  12. L2:2202 weakly excites
  13. L1:478 weakly suppresses
  14. L5:1925 weakly suppresses
  15. L5:2139 weakly excites
  16. L2:2059 weakly suppresses
  17. L5:953 weakly excites
  18. L3:672 weakly excites
  19. L1:2774 weakly suppresses
  20. L7:1590 weakly suppresses
  21. L8:2 weakly excites
  22. L6:2758 weakly suppresses
  23. L1:2392 weakly suppresses
  24. L5:1829 weakly suppresses

Feeds later layers

  1. L24:1797 weakly excites IIS and Visual Studio how-to steps
  2. L19:1132 weakly suppresses
  3. L10:1340 weakly excites
  4. L25:317 weakly excites
  5. L18:838 weakly excites
  6. L13:1905 weakly excites
  7. L18:1679 weakly excites
  8. L19:740 weakly excites
  9. L18:2330 weakly excites
  10. L18:534 weakly suppresses
  11. L14:2494 weakly excites
  12. L18:832 weakly excites
  13. L13:2098 weakly suppresses
  14. L18:2141 weakly excites
  15. L17:1795 weakly excites 'and' before a second verb phrase
  16. L15:2145 weakly excites
  17. L19:1592 weakly suppresses
  18. L13:2856 weakly suppresses
  19. L15:2124 weakly excites
  20. L17:1104 weakly excites
  21. L15:508 weakly suppresses
  22. L25:334 weakly excites
  23. L16:779 weakly excites
  24. L10:1756 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 -0.74 and -0.672 values between -0.67 and -0.603 values between -0.60 and -0.549 values between -0.54 and -0.4712 values between -0.47 and -0.4119 values between -0.41 and -0.3453 values between -0.34 and -0.28121 values between -0.28 and -0.21248 values between -0.21 and -0.15420 values between -0.15 and -0.08597 values between -0.08 and -0.01798 values between -0.01 and 0.05637 values between 0.05 and 0.12448 values between 0.12 and 0.18277 values between 0.18 and 0.25133 values between 0.25 and 0.3260 values between 0.32 and 0.3836 values between 0.38 and 0.4521 values between 0.45 and 0.518 values between 0.51 and 0.583 values between 0.58 and 0.641 values between 0.64 and 0.711 values between 1.82 and 1.89offnegativepositiveoften

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. isons
  2. typeName
  3. icy
  4. ison
  5. typename
  6. ;a
  7. ATA
  8. 重中之重
  9. _;⏎
  10. ints

Suppresses

  1. mitt
  2. 要说
  3. REET
  4. 賴
  5. ään
  6. หา
  7. imdi
  8. え
  9. 要在
  10. Speaker