Neuron Explorer

Qwen3-0.6B

'all' in 'all over the world'

A neuron in layer 24 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 'all' when it appears in phrases like 'all over the world' or 'all around Europe'. Turning it up made the model repeat 'all over the world' again and again, which matches what it responds to.

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

    ens the knowledge of science students by taking them to places that they otherwise wouldn’t be able to go. Students can see how science affects the daily lives of people all over the world and introduces memorable case studies. These are concepts which are difficult to

  2. #2 peak

    upa in Poolesville, Maryland to the Mahabodhi Stupa in Stokes Valley, New Zealand, stupas and buddhist temples are being built all over the world.⏎The Kharma

  3. #3 peak

    At least, that’s what we thought we would be doing. Continue reading →⏎Black pepper and its flavoring companion, salt, can be found in kitchens all around the world. But did you know that black pepper is not only the world

  4. #4 peak

    has just shy of 1,000 satellites in low earth orbit that are helping to provide a connection to locations all across Canada and the Northern US. These satellites work by connecting to "earth stations

  5. #5 peak

    in quality, liked the idea of using steps and paid the inventor a small fortune to install them in his palace. Soon, steps became popular all around Europe, since France was the trend setter then. The new invention was most

  6. #6 peak

    in-person tutor can be fairly expensive, an online tutor is generally inexpensive. These tutors are located all around the world, so they are available at all times of the night or day

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 an open circuit is an error, and an error is an error? What would an error in an open circuit define as? What would an error in a circuit diagram initially be? What would an error in an electrical circuit initially be

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

    Web: http://www.tandfonline.com/pricing/journal/VECE20|⏎When requesting a correction, please mention this item's handle: RePEc:taf:jeduce:v:42:y:20

  2. #2 peak

    epidemics of ‘fusarium ear blight’ (FEB), scientists in the UK and China have projected. Bruce Fitt from the University of Hertfordshire in Hatfield, UK, and his teammates forecast levels of the disease

  3. #3 peak

    biophysical forces in the extracellular environment may play an important role.⏎Not of Divided Mind⏎19.01.2017 | Hertie

  4. #4 peak

    15, Number 2—February 2009⏎Reemergence of Human and Animal Brucellosis, Bulgaria⏎B

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. L23:1361 excites
  2. L18:598 excites
  3. L22:2630 weakly suppresses
  4. L19:1305 weakly excites unclear: verbs after a noun or 'that'
  5. L22:2062 weakly suppresses
  6. L20:762 weakly suppresses
  7. L23:1368 weakly excites
  8. L15:1014 weakly suppresses
  9. L20:1789 weakly suppresses
  10. L16:2090 weakly excites
  11. L22:1998 weakly suppresses
  12. L14:1371 weakly excites unclear: document ends and instruction verbs
  13. L15:822 weakly suppresses
  14. L18:569 weakly excites
  15. L20:3050 weakly suppresses
  16. L13:1647 weakly excites
  17. L16:999 weakly excites
  18. L3:2753 weakly excites
  19. L21:794 weakly suppresses
  20. L19:1726 weakly suppresses
  21. L2:2949 weakly suppresses
  22. L7:498 weakly excites
  23. L8:2328 weakly suppresses
  24. L20:1972 weakly excites 'of' in 'of course'

Feeds later layers

  1. L27:769 strongly excites
  2. L25:1241 strongly excites
  3. L26:684 excites
  4. L26:911 excites
  5. L25:596 weakly suppresses words right before 'all' or 'in'
  6. L26:2081 weakly excites
  7. L26:631 weakly excites
  8. L26:849 weakly excites
  9. L26:1970 weakly excites
  10. L25:1139 weakly suppresses
  11. L25:576 weakly suppresses
  12. L26:418 weakly suppresses
  13. L27:3026 weakly excites
  14. L26:1995 weakly suppresses
  15. L25:1437 weakly excites
  16. L27:1795 weakly suppresses
  17. L25:734 weakly suppresses
  18. L25:2702 weakly excites
  19. L26:2448 weakly suppresses
  20. L26:2402 weakly excites
  21. L25:1132 weakly excites
  22. L27:320 weakly excites
  23. L26:144 weakly excites
  24. L25:2607 weakly 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 -17.41 and -16.181 values between -16.18 and -14.941 values between -13.71 and -12.482 values between -12.48 and -11.252 values between -11.25 and -10.024 values between -8.79 and -7.561 values between -7.56 and -6.334 values between -6.33 and -5.0914 values between -5.09 and -3.8623 values between -3.86 and -2.6350 values between -2.63 and -1.401034 values between -1.40 and -0.172301 values between -0.17 and 1.06135 values between 1.06 and 2.2988 values between 2.29 and 3.5369 values between 3.53 and 4.7650 values between 4.76 and 5.9924 values between 5.99 and 7.2225 values between 7.22 and 8.4522 values between 8.45 and 9.6814 values between 9.68 and 10.9110 values between 10.91 and 12.147 values between 12.14 and 13.388 values between 13.38 and 14.612 values between 14.61 and 15.842 values between 15.84 and 17.072 values between 17.07 and 18.301 values between 18.30 and 19.533 values between 20.76 and 21.991 values between 21.99 and 23.233 values between 23.23 and 24.462 values between 24.46 and 25.691 values between 26.92 and 28.151 values between 30.61 and 31.84offnegativepositiveoften

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. all
  2. 所有
  3. all
  4. (all
  5. 所有的
  6. All
  7. All
  8. →all
  9. .all
  10. ALL

Suppresses

  1. ignant
  2. Winner
  3. 随处可见
  4. heets
  5. ですし
  6. 什么事情
  7. 万余元
  8. 才知道
  9. dyn
  10. iversary