Neuron Explorer

Qwen3-0.6B

commas inside lists of names

A neuron in layer 10 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 one item in a list, such as a list of places, names, or things, before the next item follows. Turning it up did not produce a clear change in topic or style.

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

    the First Nation of Na-Cho Nyak Dun, the Tetlit Gwich’in Council, Tr’ondëk Hwëch’in, and the Vuntut Gwitchin First Nation, hold traditional territories in the

  2. #2 peak

    published in Journal of the American Chemical Society and can be read online.<|endoftext|>Reproductive prints—copies specifically made to replicate drawings or paintings—were collected, studied, and enjoyed widely in the eighteenth century. Interest in collecting reproductive prints stimulated new

  3. #3 peak

    very motivation for departing the Union was their objection to federal authority. On February 4, 1861, representatives from the seven states Alabama, Florida, Georgia, Louisiana, Mississippi, South Carolina, and Texas-that had already seced

  4. #4 peak

    ponder what kind of evolutionary changes might occur due to the impact of climate change.⏎Thanks to our Patrons Cristina Magistrali, Toren Wallengren, Eric Huffman, Julia Casey, Colton Siefker, Daniel Wenger, Mat

  5. #5 peak

    Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation; with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license is included in

  6. #6 peak

    . To guarantee Southerners their much-desired states' rights, the federal government had no authority to levy protective tariffs, make internal improvements, or overrule state court decisions, while states had the right to

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: I should do this, so I do, and I have, I will, and I will do, and I will do something, and I will do this, and I will do something, and I will do something, and I will

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

    agency that works with youth and that has a mission of supporting early, preventative substance abuse.⏎This curriculum is intended to be facilitated by a health or science teacher, mental health worker, law enforcement officer, or other youth services provider in a school or

  2. #2 peak

    confidential information in response to an unsolicited e-mail, text message or phone call, no matter who the source supposedly is.⏎3. Don

  3. #3 peak

    2023 by officers from Thailand’s Natural Resources and Environmental Crime Division highlighted ongoing concerns about the illegal macaque market in the Mekong region, heightened worries about the possibility of another pandemic

  4. #4 peak

    . (This also happens to be the opinion of the FDA, public health officials, and most dietitians and doctors). Raw – aka unpaste

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. L8:1605 strongly excites
  2. L5:3012 suppresses
  3. L8:250 suppresses
  4. L5:675 excites
  5. L2:1154 suppresses
  6. L7:270 suppresses
  7. L8:1238 suppresses
  8. L3:487 suppresses
  9. L3:758 excites
  10. L7:403 excites
  11. L2:66 excites
  12. L6:1230 excites
  13. L9:92 excites "and" or "or" joining two items
  14. L9:77 excites commas in citations and name lists
  15. L5:49 suppresses
  16. L6:620 suppresses "and" in lists of three or more
  17. L7:65 excites
  18. L5:548 suppresses
  19. L3:392 suppresses
  20. L3:389 suppresses
  21. L9:2302 suppresses
  22. L7:257 weakly excites
  23. L3:661 weakly excites
  24. L5:51 weakly suppresses

Feeds later layers

  1. L13:1066 strongly suppresses
  2. L15:1839 strongly suppresses
  3. L14:1746 strongly suppresses
  4. L13:1296 strongly suppresses
  5. L16:2432 strongly suppresses
  6. L16:2577 suppresses
  7. L17:2221 suppresses
  8. L12:1235 suppresses commas inside numbers and lists
  9. L11:876 excites 'other' after 'among' in lists
  10. L15:2277 excites
  11. L14:1116 excites 'and' inside lists
  12. L16:1145 suppresses commas inside short lists
  13. L12:501 suppresses
  14. L15:1213 suppresses
  15. L13:926 suppresses
  16. L19:1369 suppresses
  17. L17:919 suppresses
  18. L13:557 excites
  19. L19:2416 suppresses
  20. L13:497 excites
  21. L18:1394 suppresses
  22. L14:816 suppresses
  23. L16:1766 suppresses
  24. L12:1756 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 -11.31 and -10.982 values between -10.65 and -10.322 values between -10.32 and -9.983 values between -9.32 and -8.991 values between -8.99 and -8.661 values between -7.99 and -7.662 values between -6.99 and -6.662 values between -6.00 and -5.673 values between -5.67 and -5.333 values between -5.33 and -5.003 values between -5.00 and -4.675 values between -4.67 and -4.346 values between -4.34 and -4.005 values between -4.00 and -3.679 values between -3.67 and -3.345 values between -3.34 and -3.0110 values between -3.01 and -2.686 values between -2.68 and -2.3415 values between -2.34 and -2.018 values between -2.01 and -1.6814 values between -1.68 and -1.3523 values between -1.35 and -1.0128 values between -1.01 and -0.6850 values between -0.68 and -0.3590 values between -0.35 and -0.021325 values between -0.02 and 0.322001 values between 0.32 and 0.65283 values between 0.65 and 0.981 values between 0.98 and 1.311 values between 1.64 and 1.98offnegativepositiveoften

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. '''⏎
  3. elic
  4. .exceptions
  5. 。「
  6. ogg
  7. economic
  8. '''
  9. geb
  10. '''⏎

Suppresses

  1. anda
  2. ryo
  3. etc
  4. arie
  5. olta
  6. 让您
  7. warts
  8. าม
  9. afen
  10. raining