Neuron Explorer

Qwen3-0.6B

unclear: coordinating words in lists

A neuron in layer 17 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

The neuron is hard to pin down. It sometimes lights up on commas and 'or' or 'and' that join options in a list, but also on random words like a digit or a bracketed term. Turning it up made the model repeat 'and' and 'the' over and over.

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

    /or share your photos online.⏎A lot of online tutorials make it sound sort of scary, but it’s not. If you’re off a few (or 30) seconds or your water’s a couple degrees off temperature, you’re NOT

  2. #2 peak

    data:⏎We will work out the number of lego bricks needed in several step using a sequencing algorithm: An algorithm consisting of instructions that are carried out (performed) one after another.⏎We will use the following steps:⏎- Convert all measurement

  3. #3 peak

    this date remains on this date and is not moved to the following Monday. There has been much dispute in the past on whether weekends were on Friday and Saturday or Saturday and Sunday, so Sunday was established as the only day of rest and

  4. #4 peak

    closer zooms, you have more precision in placing the Measure tool end points.⏎If you want your measurements in strict alignment on the drawing’s vertical or horizontal (X or Y) axes, you might want to use the Horizontal measure or Vertical measure

  5. #5 peak

    ota? Also referred to as gut flora, it’s the collection of microorganisms found at any given moment in the gut – the small intestine and large intestine (colon). Our microbiota is in constant flux due to the life-cycle of these micro

  6. #6 peak

    ’ theory on perception, the two philosophies bear some similarities on duality. For instance, they both believe in the existence of a body and a soul (mind).⏎There exist two different realities for every individual that lives; that which is physical

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: and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and

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

    based on the total amount of cadmium – and not the amount extracted in a lab solution – is more health protective for children, more enforceable for government regulators, and less expensive

  2. #2 peak

    [SVG] How to draw a rectangle?⏎<svg height="210" width="500"> <polygon points="200,10 250,190 160,210

  3. #3 peak

    a major worldwide public health issue, resulting in over 13 million annual fatalities. The rise of novel illnesses, the resurgence of diseases that were previously under control, and the development of antibiotic resistance are all being influenced by changes in society, technology

  4. #4 peak

    of the human body, and as a result, dermatologists typically offer a wide variety of treatments for it.⏎Whether you wish to correct wrinkles, crepey skin, or razor bumps, a board certified dermatologist or dermasurgeon may help

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. L16:473 strongly excites
  2. L16:1119 strongly suppresses
  3. L15:2054 strongly suppresses
  4. L15:2080 suppresses
  5. L16:774 suppresses
  6. L15:2277 suppresses
  7. L16:2183 excites
  8. L15:2737 suppresses
  9. L16:1141 excites
  10. L16:521 excites
  11. L15:1839 excites
  12. L15:1699 suppresses
  13. L16:2992 suppresses
  14. L16:565 excites
  15. L15:2440 excites word before 'and' or 'or' in pairs
  16. L15:60 suppresses
  17. L16:1928 suppresses
  18. L16:2922 excites
  19. L15:2912 excites
  20. L15:389 suppresses
  21. L16:2703 excites
  22. L15:989 suppresses
  23. L16:2649 suppresses
  24. L15:1078 excites

Feeds later layers

  1. L19:164 strongly suppresses
  2. L19:1605 excites
  3. L18:1207 excites
  4. L18:2341 excites comma before an added clause
  5. L20:1009 excites
  6. L21:393 excites unclear: relative 'that' and list items
  7. L26:347 excites the 're' in self-reliant and similar
  8. L18:851 excites
  9. L22:632 excites
  10. L20:610 excites
  11. L19:867 excites
  12. L25:6 suppresses
  13. L19:1414 excites
  14. L27:289 excites
  15. L23:379 excites
  16. L20:560 excites
  17. L18:619 excites
  18. L24:205 excites
  19. L19:1092 excites
  20. L27:358 excites
  21. L18:120 suppresses
  22. L18:1413 excites
  23. L19:1369 excites
  24. L19:1245 excites comma opening a list of parallel items

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 -9.33 and -8.991 values between -7.63 and -7.281 values between -7.28 and -6.941 values between -6.94 and -6.604 values between -6.60 and -6.262 values between -5.92 and -5.584 values between -5.58 and -5.242 values between -5.24 and -4.908 values between -4.90 and -4.564 values between -4.56 and -4.226 values between -4.22 and -3.889 values between -3.88 and -3.5412 values between -3.54 and -3.2012 values between -3.20 and -2.8610 values between -2.86 and -2.5216 values between -2.52 and -2.1826 values between -2.18 and -1.8333 values between -1.83 and -1.4965 values between -1.49 and -1.1588 values between -1.15 and -0.81177 values between -0.81 and -0.47443 values between -0.47 and -0.131485 values between -0.13 and 0.21714 values between 0.21 and 0.55383 values between 0.55 and 0.89194 values between 0.89 and 1.2396 values between 1.23 and 1.5757 values between 1.57 and 1.9120 values between 1.91 and 2.2515 values between 2.25 and 2.5913 values between 2.59 and 2.932 values between 2.93 and 3.281 values between 3.28 and 3.622 values between 3.62 and 3.961 values between 3.96 and 4.30offnegativepositiveoften

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. 以及
  2. 或者
  3. and
  4. 性和
  5. および
  6. 和社会
  7. 或者是
  8. 和支持
  9. 文化和
  10. そして

Suppresses

  1. anker
  2. ervo
  3. 眄
  4. quer
  5. ву
  6. gett
  7. own
  8. ropol
  9. rella
  10. zar