Neuron Explorer

Qwen3-0.6B

"however" as a mid-sentence aside

A neuron in layer 9 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 "however" when it sits inside a sentence, set off by commas after a subject or phrase, like "Ants, however, are attracted". It was quiet on other words, and changing it did not produce a clear shared theme in the generated text.

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

    risk of birth defects and intellectual impairments. Alcohol also contributes to death and disability through accidents and injuries, assault, violence, homicide and suicide. It is, however, one of the

  2. #2 peak

    workers to be able to operate tractors and other farm machinery would be increased from 20 hours to 90 hours. Children enrolled in the program, however, could

  3. #3 peak

    idacloprid when caring for the gardenia.⏎A heavy sooty mold infestation may infect the whitefly or its larvae. Ants, however, are attracted

  4. #4 peak

    the Kennedy-Johnson administrations thought about nuclear war and nuclear weapons use in European and intercontinental military operations. The theater and strategic war plans that they approved, however, remain classified. Yet, basic planning concepts and nuclear targeting options in U.S

  5. #5 peak

    actively in the circle of expatriate modernists that included luminaries such as Gertrude Stein, Ezra Pound, and Ernest Hemingway. Fitzgerald, however, did not participate in the experiments of style and form that characterized most of the

  6. #6 peak

    of California, Los Angeles, who was not involved in the research. “And it does provide some evidence for a potential mechanism of treatment.”⏎Some researchers caution, however, that the findings

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 following is a collection of posts on the following topics: the following are the following topics: the following are the following topics: the following are the following topics: the following are the following topics: the following are the following topics: the

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

    professional health advice. This content does not provide dosage information, format recommendations, toxicity levels, or possible interactions with prescription drugs. Accordingly, this information should be used only under the

  2. #2 peak

    of new digital content poses a real challenge for archivists. Hard disks are expensive and require a constant supply of electricity, while even the best 'no-power' archiving materials such as magnetic tape degrade within a decade. This is a growing

  3. #3 peak

    tight budget. In several places, the ceiling lets through water from the grey Rhode Island sky.⏎The youngest group of children at Federal Hill

  4. #4 peak

    , looking for ideas that, for whatever reason, resonate with you?⏎Will you resist distraction while you read this? Or are you likely to have your attention pulled away - to jump to another open tab, or simply switch your focus to what’s

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. L8:140 excites
  2. L6:2630 suppresses
  3. L7:32 excites
  4. L8:2298 suppresses
  5. L5:2146 excites
  6. L7:2280 suppresses
  7. L8:2296 suppresses
  8. L8:2646 excites
  9. L5:2502 weakly excites
  10. L4:9 weakly excites
  11. L8:1395 weakly excites
  12. L7:612 weakly suppresses
  13. L7:2779 weakly suppresses
  14. L6:1461 weakly suppresses
  15. L8:3036 weakly excites
  16. L7:443 weakly suppresses 'not' or 'only' setting up a contrast
  17. L7:2653 weakly suppresses
  18. L6:2457 weakly excites
  19. L5:2775 weakly suppresses
  20. L7:1387 weakly excites
  21. L6:2288 weakly suppresses
  22. L8:2191 weakly suppresses
  23. L7:2169 weakly suppresses
  24. L5:6 weakly suppresses

Feeds later layers

  1. L11:1922 excites
  2. L12:3025 excites
  3. L11:2178 excites
  4. L12:2154 excites
  5. L12:20 suppresses
  6. L12:877 excites
  7. L11:106 suppresses
  8. L10:449 excites
  9. L11:1289 excites
  10. L13:1233 excites
  11. L10:2301 excites
  12. L11:2911 excites
  13. L12:861 excites
  14. L14:3007 excites
  15. L17:2798 excites
  16. L11:1933 weakly suppresses
  17. L11:1653 weakly excites
  18. L11:158 weakly suppresses
  19. L12:2339 weakly excites
  20. L15:1243 weakly excites
  21. L10:2215 weakly excites
  22. L15:1725 weakly excites
  23. L11:2730 weakly excites
  24. L10:673 weakly 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.

1 values between -3.22 and -3.021 values between -2.82 and -2.622 values between -2.62 and -2.421 values between -2.23 and -2.035 values between -2.03 and -1.832 values between -1.83 and -1.632 values between -1.63 and -1.4311 values between -1.43 and -1.242 values between -1.24 and -1.047 values between -1.04 and -0.8423 values between -0.84 and -0.64120 values between -0.64 and -0.44605 values between -0.44 and -0.251297 values between -0.25 and -0.051265 values between -0.05 and 0.15282 values between 0.15 and 0.35117 values between 0.35 and 0.5553 values between 0.55 and 0.7438 values between 0.74 and 0.9416 values between 0.94 and 1.1420 values between 1.14 and 1.3410 values between 1.34 and 1.547 values between 1.54 and 1.736 values between 1.73 and 1.931 values between 1.93 and 2.135 values between 2.13 and 2.333 values between 2.33 and 2.532 values between 2.92 and 3.122 values between 3.32 and 3.522 values between 4.51 and 4.70offnegativepositiveoften

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. opes
  2. answered
  3. ace
  4. don
  5. oid
  6. Don
  7. sur
  8. ollapse
  9. цу
  10. üm

Suppresses

  1. bourne
  2. .Append
  3. 书中
  4. _CAPTURE
  5. irectory
  6. 标的
  7. 一则
  8. receive
  9. 史诗
  10. atcher