Neuron Explorer

Qwen3-0.6B

comma after a filler word

A neuron in layer 6 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 a comma that comes right after a small phrase like "well,", "however," or "oh,". It is quiet on words like "to begin with" and "more tragically". Turning it up or down gave off-topic text, so it does not tell us much about what it does.

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

    qualities that make it truly distinctive. The second character drawings are stronger structurally and easier to understand, but tend to look more regularized, streamlined and, well, normal.⏎The point is to be aware of the surprising shapes, lines and proportions

  2. #2 peak

    many specific tick species within each of these categories. Some ticks are found across the United States. Others are limited to specific regions of the country. Thankfully, however, most ticks are fairly easy to identify.⏎Genus & Species⏎& Conditions Ca

  3. #3 peak

    when some 5 million people were living with the disease. Since then, AIDS has killed more than 20 million people around the world and, this year, 40 million people are living with HIV/AIDS. In sub-Saharan Africa alone

  4. #4 peak

    clusters; they are among the biggest coherent structures in the Universe (what’s bigger? Superclusters, which are clusters of clusters!). Imagine a collection of, oh, say, a thousand galaxies the size of the Milky Way. The sky inside one

  5. #5 peak

    what appears to be an image of, well, absolutely nothing: a fascinating image of what appears to be a hole in the fabric

  6. #6 peak

    makes them less worthy of attention, concern, and research money than, say, childhood obesity, which as we all know is public healthy enemy #1. (

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: if $ x $ is real, $ x $ is a complex number, then, what is the value of the expression? Alright, let's start by, let's think... the first thing, the first part of the problem. the

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

    can the climate generation do?⏎To begin with, business as usual is not an option. These young people are green entrepreneurs and

  2. #2 peak

    's first pictures from the DC-8, posted to ESA's ATV blog, revealed a spray of fireworks similar to those seen during Mir's fall. More tragically, they also recalled what witnesses saw during the shuttle Columbia's breakup five years ago

  3. #3 peak

    “Classroom flipping” was a hit, resulting in wide smiles on the murky faces of students and revival of new spirit of teaching in teachers.⏎· To begin with, classroom flipping requires that you reach to

  4. #4 peak

    is a plant that is grown, among other things, for consumption and to make users ‘high’. Thus, we can safely call marijuana a drug.⏎To start with, marijuana, which can come from the cannabis Indica or Sativa species,

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. L4:1441 excites
  2. L4:169 suppresses
  3. L3:682 excites
  4. L0:828 weakly suppresses
  5. L4:486 weakly excites
  6. L5:2641 weakly excites unclear: scattered word fragments
  7. L2:1254 weakly excites
  8. L0:538 weakly excites
  9. L2:2211 weakly excites
  10. L4:1444 weakly suppresses
  11. L4:592 weakly excites
  12. L5:177 weakly excites
  13. L5:321 weakly excites
  14. L1:239 weakly excites
  15. L5:1755 weakly excites
  16. L4:2306 weakly excites
  17. L5:2309 weakly suppresses
  18. L3:774 weakly suppresses
  19. L1:2176 weakly excites
  20. L3:2419 weakly excites
  21. L3:1249 weakly suppresses
  22. L5:2622 weakly excites
  23. L2:2658 weakly excites
  24. L5:1039 weakly suppresses

Feeds later layers

  1. L8:27 strongly suppresses
  2. L8:79 strongly suppresses
  3. L7:7 strongly suppresses
  4. L16:269 suppresses
  5. L7:8 suppresses
  6. L13:3 suppresses
  7. L17:70 suppresses
  8. L18:513 suppresses
  9. L14:417 suppresses
  10. L7:135 excites 'of' after 'issue' in citations
  11. L17:432 suppresses
  12. L14:118 suppresses
  13. L16:558 suppresses
  14. L19:204 suppresses
  15. L12:50 suppresses
  16. L7:41 excites
  17. L14:830 excites
  18. L9:330 excites gap where a number was removed
  19. L18:126 suppresses comma joining a full clause
  20. L9:141 suppresses
  21. L25:16 suppresses
  22. L7:233 suppresses
  23. L20:317 suppresses
  24. L16:180 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 -4.32 and -4.151 values between -2.48 and -2.311 values between -2.15 and -1.982 values between -1.64 and -1.481 values between -1.48 and -1.311 values between -1.14 and -0.973 values between -0.97 and -0.812 values between -0.81 and -0.646 values between -0.64 and -0.4772 values between -0.47 and -0.30633 values between -0.30 and -0.141957 values between -0.14 and 0.031046 values between 0.03 and 0.20137 values between 0.20 and 0.3719 values between 0.37 and 0.536 values between 0.53 and 0.702 values between 0.70 and 0.874 values between 0.87 and 1.033 values between 1.03 and 1.204 values between 1.20 and 1.373 values between 1.37 and 1.542 values between 1.54 and 1.701 values between 1.87 and 2.041 values between 2.21 and 2.37offnegativepositiveoften

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. <?=$
  2. phia
  3. eya
  4. isme
  5. ombat
  6. ignKey
  7. <?=
  8. gium
  9. ters
  10. ziej

Suppresses

  1. Mi
  2. -fe
  3. env
  4. Ent
  5. Growing
  6. fe
  7. 认
  8. 转载请
  9. fc
  10. Fin