Neuron Explorer

Qwen3-0.6B

commas inside lists of nouns

A neuron in layer 13 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 commas and dashes that separate items in a list of nouns, like 'resources, natural disasters' or bullet points of symptoms. It is quiet or negative inside medical symptom lists. Turning it up gave text that did not clearly follow the list idea.

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

    00<|endoftext|>The reasons and urgency that lead a group of people to just up and leave a place are varied: obsolete technology, depleted resources, natural or man-made disasters. But whatever the case, there will always be people

  2. #2 peak

    ised in sociology to spell out the prevailing patterns of behavior and belief among members of your society or community, including these kinds of factors seeing that language, religious beliefs, relatives practices, monetary systems, and belief and value systems.⏎Internet dating Culture:

  3. #3 peak

    only need the services of a general dentist, there are many other issues that require specialist services. An endodontist provides treatment for:⏎- Chipped teeth⏎- Broken teeth⏎- Infected teeth⏎- Cracked teeth⏎- Infl

  4. #4 peak

    of distinct characteristics that this map aspires to signify. There are several forms of maps that attempt to symbolize particular things. Maps can display political borders, inhabitants, bodily characteristics, all-natural solutions, highways, environments, height (topography

  5. #5 peak

    under control so you can react to river hazards, obstacles and other boats.⏎Know and be aware of river signs, both natural and man-made, such as holes, wrap rocks, under cuts, rock sieves, horizon lines across the river --

  6. #6 peak

    1 percent in Hispanics).These racial/ethnic differences in the use of smoking cessation remained significant even after controlling for various other factors (for example, health insurance coverage, or socio-economics status of smokers).⏎“The good news is the number of smokers

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: it's been a while since I've had a chance to read it? Or maybe not? Or not? Maybe not? Maybe not? Or not? Or not? Maybe not? Maybe not? Or not? Maybe not? Maybe not

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

    starts.⏎Stroke signs and symptoms include:⏎Trouble talking and understanding what others are saying. You may get upset, talk blurred or have trouble understanding speech.⏎Paralysis or numbness of the face, arm, or leg may occur. You may

  2. #2 peak

    is a slowly progressing, degenerative disease that is usually associated with the following symptoms, all of which result from the loss of dopamine-producing brain cells:⏎- tremor or trembling of the arms, jaw,

  3. #3 peak

    signs or symptoms of strep throat, see your doctor for prompt testing and treatment.⏎Signs and symptoms of strep throat can include:⏎- Throat pain that usually comes on quickly⏎- Painful swallowing⏎- Red and swollen tonsils

  4. #4 peak

    with the Zero Waste Concept⏎Adapt to the 3 R’s of Waste Hierarchy⏎The 3 R’s of the waste hierarchy are,⏎- Reduce – It best to reduce and minimize buying unnecessary items.⏎- Reuse – Consider re

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. L12:1756 excites
  2. L12:365 suppresses
  3. L12:1252 suppresses
  4. L12:690 excites
  5. L12:2071 excites
  6. L12:990 suppresses
  7. L12:314 excites
  8. L12:2600 excites
  9. L12:660 excites
  10. L12:94 excites
  11. L12:355 excites
  12. L12:2957 excites
  13. L12:2066 excites
  14. L12:152 excites
  15. L12:2204 suppresses
  16. L8:986 suppresses
  17. L10:122 excites
  18. L11:729 excites
  19. L11:1251 excites
  20. L12:1379 weakly excites
  21. L12:2286 weakly excites
  22. L12:1945 weakly suppresses
  23. L12:2981 weakly suppresses
  24. L12:185 weakly suppresses

Feeds later layers

  1. L17:518 excites
  2. L16:2251 suppresses
  3. L19:1820 suppresses
  4. L18:2118 suppresses
  5. L16:1572 suppresses
  6. L17:84 suppresses
  7. L17:1117 excites
  8. L17:997 weakly suppresses
  9. L16:300 weakly excites commas inside lists of nouns
  10. L16:1427 weakly excites
  11. L18:101 weakly suppresses
  12. L14:1082 weakly suppresses
  13. L22:368 weakly excites
  14. L15:1798 weakly suppresses
  15. L18:2078 weakly excites
  16. L17:173 weakly suppresses opening bracket that adds a synonym
  17. L14:1386 weakly suppresses
  18. L16:1977 weakly suppresses
  19. L15:1064 weakly excites
  20. L20:277 weakly suppresses
  21. L14:1321 weakly suppresses
  22. L15:2314 weakly excites
  23. L21:76 weakly suppresses
  24. L18:2021 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 -3.09 and -2.901 values between -2.72 and -2.531 values between -2.34 and -2.151 values between -1.96 and -1.775 values between -1.77 and -1.583 values between -1.58 and -1.392 values between -1.39 and -1.205 values between -1.20 and -1.0118 values between -1.01 and -0.8236 values between -0.82 and -0.6351 values between -0.63 and -0.44132 values between -0.44 and -0.25559 values between -0.25 and -0.062399 values between -0.06 and 0.13491 values between 0.13 and 0.32101 values between 0.32 and 0.5035 values between 0.50 and 0.6920 values between 0.69 and 0.8813 values between 0.88 and 1.074 values between 1.07 and 1.269 values between 1.26 and 1.456 values between 1.45 and 1.645 values between 1.64 and 1.831 values between 1.83 and 2.022 values between 2.02 and 2.211 values between 2.40 and 2.592 values between 2.59 and 2.781 values between 2.97 and 3.161 values between 3.16 and 3.341 values between 3.53 and 3.721 values between 4.29 and 4.48offnegativepositiveoften

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. ': ⏎
  4. 摊
  5. ':⏎⏎
  6. mond
  7. __':⏎
  8. __('
  9. _='
  10. ']:⏎

Suppresses

  1. бо
  2. yper
  3. */,
  4. porta
  5. enan
  6. plib
  7. 或将
  8. »,
  9. }`,
  10. бор