Neuron Explorer

Qwen3-0.6B

'or' in lists of alternatives

A neuron in layer 15 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 the word 'or' (and similar words like 'even') when it joins two items in a list of options, such as 'accuracy or completeness'. Turning it up made the model produce loose lists of nouns, and turning it down made it repeat phrases.

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

    they explicitly disclaim, any express or implied representations or warranties of any kind, including, without limitation, representations and warranties as to the functionality of the translation feature or the accuracy or completeness of the translations.⏎Translations are not retained in our system.

  2. #2 peak

    diagnose the cause of failing paint on a building exterior or interior⏎InspectAPedia tolerates no conflicts of interest. We have no relationship with advertisers, products, or services discussed at this website.⏎Paint failure causation, diagnosis, cure: this

  3. #3 peak

    The background is additionally the sight in the picture since it adds rate of interest or drama. In portrait photography the history can be virtually anything – the pet, blossom or rock.⏎One more prominent sort of portrait digital photography is landscape photography. Landscape photography

  4. #4 peak

    over and over. Simply print and students are ready to engage!⏎***Just a friendly reminder that this product is for A SINGLE USER ONLY. That means that copying or sharing any part of this product with another teacher, grade level, school, department

  5. #5 peak

    not philosophical, speculative, or ethical elements; no⏎elaborate ritual or priestly hierarchy; no saints, martyrs, or even a⏎founder; no scripture or exegesis; no interest in education

  6. #6 peak

    has a right to vote on some corporate matters. A typical limited partnership agreement may provide that a majority vote of the limited partners is necessary for the sale of assets or to remove a general partner. The partnership agreement determines whether the limited partners can vote

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 about the, what about, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as, as

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

    ), or to adelmnops in any order:⏎s - endoplasms⏎All words formed from endoplasm by changing one letter⏎Browse words starting with endoplasm by next letter<|endoftext|>- Historic Sites⏎The Winter Soldiers,⏎December

  2. #2 peak

    agrams for focus found in this word list.⏎List shorter words within focus, sorted by length⏎All words formed from focus by changing one letter⏎Browse words starting with focus by next letter<|endoftext|>I was listening to Brooke Gladstone's ON THE MEDIA

  3. #3 peak

    and hence life could exist.⏎Kepler does its work by noting the dimming of its target star as a planet or planets pass between the star and the telescope. Scientists are then able to note both the approximate sizes of

  4. #4 peak

    Rick Miller⏎Classic stamps lie near the heart of U.S. stamp collecting. A good place to start or to finish is at the beginning.⏎The first U.S. general-issue postage stamps were issued July 1,

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. L14:1571 strongly suppresses
  2. L14:1184 suppresses
  3. L14:1345 excites
  4. L14:1500 suppresses
  5. L14:516 excites
  6. L12:1094 excites
  7. L14:1174 excites
  8. L12:2334 suppresses
  9. L13:562 suppresses
  10. L13:760 excites
  11. L13:1795 excites second item in a parallel pair or list
  12. L11:125 excites
  13. L14:3061 suppresses
  14. L14:613 suppresses
  15. L14:3034 excites
  16. L12:1720 suppresses
  17. L13:899 suppresses
  18. L14:1213 excites
  19. L12:1280 suppresses
  20. L14:2328 suppresses
  21. L14:1615 excites
  22. L12:152 excites
  23. L13:1549 excites
  24. L14:1433 suppresses

Feeds later layers

  1. L18:2362 strongly suppresses
  2. L25:1270 suppresses
  3. L20:1334 suppresses
  4. L19:3001 excites
  5. L17:1897 suppresses
  6. L23:223 suppresses
  7. L16:2329 excites
  8. L19:289 excites
  9. L16:774 excites
  10. L27:376 suppresses
  11. L17:2891 suppresses
  12. L16:2992 suppresses
  13. L20:1599 suppresses
  14. L16:2649 suppresses
  15. L26:2473 suppresses
  16. L26:779 suppresses
  17. L25:41 excites
  18. L16:935 excites
  19. L26:492 suppresses
  20. L16:850 excites
  21. L16:2442 suppresses
  22. L17:1462 excites
  23. L19:1325 suppresses
  24. L23:379 excites

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.

2 values between -2.09 and -1.952 values between -1.80 and -1.666 values between -1.37 and -1.223 values between -1.22 and -1.087 values between -1.08 and -0.9319 values between -0.93 and -0.7920 values between -0.79 and -0.6442 values between -0.64 and -0.5075 values between -0.50 and -0.35193 values between -0.35 and -0.21502 values between -0.21 and -0.061807 values between -0.06 and 0.09559 values between 0.09 and 0.23254 values between 0.23 and 0.38134 values between 0.38 and 0.5295 values between 0.52 and 0.6767 values between 0.67 and 0.8122 values between 0.81 and 0.9631 values between 0.96 and 1.1017 values between 1.10 and 1.2510 values between 1.25 and 1.3914 values between 1.39 and 1.546 values between 1.54 and 1.684 values between 1.68 and 1.836 values between 1.83 and 1.971 values between 1.97 and 2.122 values between 2.12 and 2.263 values between 2.26 and 2.411 values between 2.56 and 2.702 values between 2.85 and 2.991 values between 3.43 and 3.571 values between 3.57 and 3.72offnegativepositiveoften

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. &view
  2. 并与
  3. 且
  4. ấy
  5. ided
  6. 拿了
  7. 조사
  8. &P
  9. éparation
  10. +y

Suppresses

  1. 或者
  2. or
  3. 或是
  4. 或多
  5. 或
  6. 或者是
  7. 或
  8. ||⏎
  9. 或其他
  10. 或将