Neuron Explorer

Qwen3-0.6B

'or' joining alternatives in lists

A neuron in layer 7 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 lights up on the word 'or' when it links two or more options, like 'soap or warm water' or 'launching a product or working for'. It goes quiet or negative on 'and' in similar spots. Turning it up made the model's text drift toward unrelated topics, so the steering results don't reveal much more.

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

    that’s launching a new product, or they might work for<|endoftext|>Cite This Page⏎Section III (lines 1

  2. #2 peak

    frequently felt when you begin a new exercise program, change your routine, or dramatically increase the duration or intensity of your workout. DOMS is a normal response to unusual exertion and

  3. #3 peak

    analytics together seamlessly in the same application.⏎- Learn how it runs on top of Hadoop, Mesos, standalone, or in the cloud. It can access diverse data sources such as HDFS, Cassandra

  4. #4 peak

    and lifelong learners.⏎Whenever you are stuck, need a new idea, or want to learn and grow, The 5 Elements of Effective Thinking will inspire and

  5. #5 peak

    ate equipment to heat biomass, using knowledge of controls, combustion, and firing mechanisms.⏎- Assess quality of biomass feedstock.⏎- Calculate, measure, load, or mix biomass feedstock for power generation.⏎- Manage parts and supply inventories for

  6. #6 peak

    comfortable, or sugar coated, term for racism. Racism is a practice which requires belief,

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: the problem of a thousand viruses, and the universe. The problem of a thousand viruses is the problem of a thousand worlds, but the universe, and the world, and the world, and the world, and the world, and the world

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

    - Sleeping in an awkward position⏎- Neck muscle injury at birth⏎- Burn injury⏎- Any injury that causes heavy scarring and skin or muscle shrinkage⏎- Neck muscle spasm⏎Torticollis

  2. #2 peak

    each packet and thus prevents the types of attacks that compromised WEP.⏎WPA also includes a Message Integrity Check, which is designed to prevent an attacker from altering and resending data packets. This replaces the cyclic redundancy check (CRC) that was

  3. #3 peak

    action is taken to protect the health of all Albertans.⏎You can protect yourself from all coronaviruses by:⏎- Cleaning your hands frequently using either soap and warm water or alcohol-based hand sanitizers,⏎- Covering your cough and sne

  4. #4 peak

    arms, and legs.⏎Acute Chest Syndrome: It occurs when sickling occurs in the chest. It happens suddenly when the body is under infection, fever, and

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. L3:2967 weakly excites
  2. L6:295 weakly suppresses
  3. L0:2257 weakly suppresses
  4. L5:2143 weakly excites
  5. L0:1927 weakly excites
  6. L5:2227 weakly excites
  7. L6:1376 weakly excites
  8. L3:433 weakly suppresses
  9. L4:2950 weakly suppresses
  10. L4:17 weakly excites
  11. L1:1750 weakly suppresses
  12. L3:1562 weakly suppresses
  13. L6:834 weakly suppresses
  14. L5:233 weakly suppresses
  15. L1:2593 weakly excites
  16. L6:647 weakly excites
  17. L1:2519 weakly suppresses
  18. L5:264 weakly suppresses
  19. L3:726 weakly suppresses
  20. L5:2246 weakly suppresses
  21. L3:11 weakly suppresses
  22. L2:2875 weakly excites 'up' after show, turn and showed
  23. L2:2181 weakly suppresses
  24. L3:250 weakly excites

Feeds later layers

  1. L10:135 excites 'or' in alternative requirements
  2. L15:2935 excites
  3. L8:241 suppresses
  4. L8:1475 excites
  5. L18:2197 excites
  6. L15:1907 excites the word 'or' between alternatives
  7. L10:901 suppresses
  8. L11:273 suppresses 'and' joining clauses in instructions
  9. L11:900 excites 'or' in lists of alternatives
  10. L13:553 excites
  11. L9:1066 excites
  12. L10:633 excites
  13. L11:360 excites
  14. L12:683 suppresses
  15. L10:1053 excites
  16. L12:836 excites
  17. L11:2817 excites
  18. L15:2203 suppresses
  19. L9:352 excites
  20. L16:1367 excites
  21. L17:2166 weakly excites
  22. L11:127 weakly excites
  23. L8:623 weakly excites
  24. L12:1094 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 -5.88 and -5.651 values between -3.12 and -2.892 values between -2.66 and -2.431 values between -2.43 and -2.201 values between -2.20 and -1.973 values between -1.97 and -1.743 values between -1.51 and -1.283 values between -1.28 and -1.055 values between -1.05 and -0.824 values between -0.82 and -0.5924 values between -0.59 and -0.36377 values between -0.36 and -0.132435 values between -0.13 and 0.10885 values between 0.10 and 0.3379 values between 0.33 and 0.5650 values between 0.56 and 0.7912 values between 0.79 and 1.027 values between 1.02 and 1.258 values between 1.25 and 1.481 values between 1.48 and 1.711 values between 1.71 and 1.941 values between 1.94 and 2.171 values between 2.40 and 2.631 values between 2.63 and 2.861 values between 2.86 and 3.091 values between 3.09 and 3.32offnegativepositiveoften

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. ner
  2. 其它
  3. 其他国家
  4. other
  5. 或者其他
  6. 其他人
  7. MAD
  8. 其他
  9. 其他的
  10. _Inter

Suppresses

  1. uly
  2. vetica
  3. auty
  4. ucs
  5. aurus
  6. ROID
  7. uring
  8. igned
  9. ур
  10. uality