Neuron Explorer

Qwen3-0.6B

hand and body action verbs in instructions

A neuron in layer 8 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on physical action verbs in how-to writing, such as 'drive', 'twist', 'turn', 'shake', 'open up' and 'pinch'. These are words for a hand or body doing something to an object. Turning it up did not give writing that clearly matched this 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

    their influence in the cities designed in their own style. True to its past, Delhi even today is the center of India’s political and governmental activities.⏎When you drive through the lesser known corners the experiences are most rewarding such as Lodi Gardens at

  2. #2 peak

    in the process. This will ultimately hinder the performance of the engine.⏎2) Difficult to Turn Over⏎When you put your key into the ignition and twist it, you may find the engine is difficult to turn over or start altogether.

  3. #3 peak

    statues that had survived for thousands of years.⏎Today, the museum’s darkened interior stands as a memorial to a 21st-century cultural tragedy.⏎Stepping between pieces of Assyrian sculptures blown up by explosives, Mosul Museum director Zaid

  4. #4 peak

    . If this occurs, the freezer may still get cold, while the refrigerator will not get cold. To determine if the evaporator fan motor is defective, try turning the fan blade by hand. If the fan blade does not turn freely, replace

  5. #5 peak

    The journey starts on the station platform, with the excitement of finding your name and seat number posted outside the compartment, among a list of fellow passengers. If you turn up without a booking, you will have to take your chances in the no-h

  6. #6 peak

    the sound matching activity to your child. Shake, listen, match. You can say, “May I show you how we play?” Pick up an egg and shake it. Pick up another egg in your opposite hand and shake it. If the

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: **What if?** **Yes,** **No explanation** **Original Explanation:** **First Name:**** **Last Name:** **First Name:** **Last Name:** **Second Name:** **From:** **

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

    connecting a time-limit fuse across the trip coil terminals i.e. in parallel. This fuse should have an inverse time current characteristic

  2. #2 peak

    in the form of a chain e-mail message, it has no scientific basis. In fact, following its advice can do harm.⏎Forcing a stroke victim to sit up is never a good idea, because it can cause a drop in blood pressure

  3. #3 peak

    apart. Partner 1 holds medicine ball (6–10 pounds) and rotates through transverse plane, passing ball to Partner 2 from behind. They perform

  4. #4 peak

    booster seats until the lap-shoulder belt fits properly (typically for children ages 8-12 and over 4’9”), and ride in the back seat until they are at least 13 years old.⏎Don’t drive distracted

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. L7:1708 excites
  2. L7:204 excites
  3. L3:2845 weakly suppresses
  4. L7:302 weakly suppresses
  5. L7:1504 weakly excites
  6. L4:1847 weakly suppresses
  7. L4:2865 weakly excites
  8. L7:941 weakly suppresses
  9. L6:2824 weakly suppresses
  10. L4:1274 weakly excites word right before a term's name
  11. L5:1828 weakly excites
  12. L5:459 weakly suppresses
  13. L1:1669 weakly suppresses
  14. L6:1226 weakly suppresses
  15. L6:2968 weakly excites
  16. L0:2084 weakly suppresses
  17. L6:2660 weakly excites
  18. L6:2261 weakly excites
  19. L4:2576 weakly suppresses
  20. L2:2809 weakly excites
  21. L4:3069 weakly suppresses
  22. L6:2048 weakly suppresses
  23. L0:2660 weakly suppresses
  24. L5:278 weakly excites

Feeds later layers

  1. L13:2297 excites
  2. L10:1996 excites
  3. L10:2919 excites
  4. L9:1209 suppresses
  5. L11:770 excites
  6. L9:2027 excites
  7. L11:963 excites
  8. L10:1311 weakly excites
  9. L13:610 weakly excites
  10. L18:2988 weakly excites
  11. L12:339 weakly suppresses
  12. L14:2372 weakly suppresses
  13. L21:1394 weakly excites
  14. L21:3000 weakly excites
  15. L11:1094 weakly excites unclear: scattered web and commerce text
  16. L12:2313 weakly excites
  17. L15:2646 weakly excites
  18. L9:198 weakly excites
  19. L11:1658 weakly suppresses
  20. L11:375 weakly excites
  21. L11:3025 weakly excites
  22. L16:2598 weakly excites
  23. L16:680 weakly excites
  24. L10:2295 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 -1.27 and -1.142 values between -0.87 and -0.745 values between -0.74 and -0.613 values between -0.61 and -0.4761 values between -0.47 and -0.34381 values between -0.34 and -0.211114 values between -0.21 and -0.071586 values between -0.07 and 0.06564 values between 0.06 and 0.19132 values between 0.19 and 0.3335 values between 0.33 and 0.468 values between 0.46 and 0.594 values between 0.59 and 0.732 values between 0.73 and 0.861 values between 0.86 and 0.992 values between 0.99 and 1.131 values between 1.13 and 1.261 values between 1.39 and 1.531 values between 1.79 and 1.921 values between 1.92 and 2.061 values between 2.72 and 2.861 values between 2.99 and 3.121 values between 3.92 and 4.06offnegativepositiveoften

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. смотреть
  2. 档
  3. Palindrome
  4. 给她
  5. symmetry
  6. (interface
  7. Suff
  8. 拨
  9. paz
  10. checker

Suppresses

  1. amilies
  2. ueue
  3. família
  4. 家庭教育
  5. 天河
  6. RIES
  7. families
  8. لوم
  9. hq
  10. Toro