Neuron Explorer

Qwen3-0.6B

comma in lists before "and" or "nor"

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

The neuron lights up on a comma that comes right before the last item of a list, like "chemicals[,] and UV radiation". It is less active on the word "nor" that follows a negation. Turning it up or down made the model's writing shift topic without a clear shared 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

    and a portion of the Great Eastern Trail. The historic Civilian Conservation Corps (CCC) built many of the trails.⏎With foresight, good judgment, determination, and your help⏎we can ensure that this unique, natural area will continue to

  2. #2 peak

    methodological aspects of WWC reviews and reports.⏎The WWC is committed to ensuring that its products and services meet user needs and invites input from educators, policymakers, researchers, and members of the public. All products are available through the WWC

  3. #3 peak

    inghouse (WWC) was established in 2002 by the U.S. Department of Education's Institute of Education Sciences to provide educators, policymakers, and the public with a central and trusted source of scientific evidence of what works in

  4. #4 peak

    seek a new relationship with the continent. But in the complex world of post-colonial relationships, such a move is difficult. France retains strong economic, political, and social links with West Africa. Paris, Marseille, and Lyon are home to

  5. #5 peak

    has temples, monasteries, and priests who make offerings, meditate, and perform other rituals for their communities. One of the main ideas of Taoism

  6. #6 peak

    .<|endoftext|>Newegg.com - A great place to buy computers, computer parts, electronics, software, accessories, and DVDs online. With great prices, fast shipping, and top-rated customer service - once you know, you Newegg.⏎If you

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 is the value of the expression for the number of 100? Let me explain the work done on the problem? I will start by putting the details of the problem. The value of the expression for the number of 1

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

    for some. One worker left the following anonymous note when she left:⏎“I am going home where I shall not be obliged to rise so early in the morning, nor be dragged about by the factory bell, nor confined in a close noisy room from

  2. #2 peak

    the bright morning and the stillness of night, he has been reassured, so as to say: "Your Lord has not at all forsaken you, nor is he displeased with you." Then, he has been given the good news

  3. #3 peak

    students can’t learn if they are distracted by hunger in the classroom, nor can they<|endoftext|>It's time for another tour of international food-borne and water

  4. #4 peak

    “We don’t see everything Israel does as holy, nor what the Arab states or Palestinians do…. It’s not as if we say,

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. L0:1735 weakly excites
  2. L1:2847 weakly suppresses
  3. L0:1765 weakly excites
  4. L1:2319 weakly suppresses
  5. L4:599 weakly excites
  6. L3:301 weakly suppresses
  7. L1:268 weakly excites
  8. L3:433 weakly suppresses
  9. L4:1630 weakly suppresses
  10. L2:1150 weakly suppresses
  11. L4:1753 weakly suppresses
  12. L1:2259 weakly excites
  13. L5:1205 weakly excites
  14. L4:2021 weakly suppresses
  15. L1:2906 weakly suppresses
  16. L5:2612 weakly suppresses
  17. L1:1931 weakly excites
  18. L4:1081 weakly suppresses
  19. L0:2074 weakly suppresses
  20. L3:946 weakly suppresses
  21. L2:1154 weakly suppresses
  22. L5:1936 weakly excites
  23. L1:363 weakly excites
  24. L4:690 weakly suppresses

Feeds later layers

  1. L11:251 suppresses
  2. L16:1074 suppresses
  3. L8:70 suppresses 'and' between stacked superlatives
  4. L10:1333 weakly suppresses
  5. L8:562 weakly suppresses
  6. L11:682 weakly suppresses gap before a second number
  7. L11:406 weakly suppresses
  8. L11:1595 weakly suppresses
  9. L8:593 weakly suppresses
  10. L8:1047 weakly excites
  11. L16:2111 weakly excites
  12. L8:321 weakly suppresses
  13. L18:1386 weakly suppresses unclear: words before 'and' or 'or'
  14. L9:212 weakly excites
  15. L13:716 weakly suppresses
  16. L12:990 weakly suppresses
  17. L17:1134 weakly suppresses
  18. L7:25 weakly suppresses
  19. L12:1197 weakly suppresses
  20. L14:1502 weakly suppresses
  21. L16:300 weakly suppresses commas inside lists of nouns
  22. L12:2936 weakly suppresses
  23. L12:434 weakly excites
  24. L10:838 weakly suppresses comma in a list of paired nouns

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 -1.94 and -1.793 values between -1.79 and -1.642 values between -1.64 and -1.501 values between -1.35 and -1.2011 values between -1.20 and -1.0615 values between -1.06 and -0.918 values between -0.91 and -0.7622 values between -0.76 and -0.6127 values between -0.61 and -0.4749 values between -0.47 and -0.32297 values between -0.32 and -0.171213 values between -0.17 and -0.021840 values between -0.02 and 0.12279 values between 0.12 and 0.2781 values between 0.27 and 0.4226 values between 0.42 and 0.5711 values between 0.57 and 0.716 values between 0.71 and 0.863 values between 0.86 and 1.012 values between 1.01 and 1.152 values between 1.15 and 1.302 values between 1.30 and 1.452 values between 1.45 and 1.602 values between 1.89 and 2.041 values between 2.04 and 2.181 values between 2.92 and 3.071 values between 3.80 and 3.95offnegativepositiveoften

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. 核心技术
  2. untas
  3. ===
  4. .dispatch
  5. laughs
  6. resh
  7. ��
  8. oun
  9. nicas
  10. 咄

Suppresses

  1. 迹
  2. ⌺
  3. itories
  4. ď
  5. .RowStyle
  6. azio
  7. tributes
  8. ilities
  9. rück
  10. ilded