Neuron Explorer

Qwen3-0.6B

comma before but also in not only constructions

A neuron in layer 17 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

It fires on the comma that comes after a clause such as "not only X", right before the "but also" part that completes the contrast. Turning it down made the model repeat "instead of" and "rather than", which are the words that usually set up this kind of contrast, while turning it up gave lists of "and the same" phrasing.

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

    you must understand and apply some crucial National Electrical Code rules and regulation. Following these rules to the letter will not only assure you that your work will pass inspection, but that it will be a safe installation as well. Improperly installed recess

  2. #2 peak

    mode of literary expression, that of the Petrarchan ideal of beauty and virtue. Tarquin may then be understood, Washington stresses, not just as the villain, but as one who commits a transgressive but necessary purging of literary hegemony

  3. #3 peak

    quarter of the 19th century and the decades leading up to World War II. The era is characterized by dramatic transformations in not just technological and industrial developments, but also in social and cultural institutions, including the world of art in its various

  4. #4 peak

    unhealthy social norms and values are ignored. People come to construe healthy behavior in terms of dependency on a medical industry that constantly invents not only new cures, but new diseases for the cures it already possesses (Blech,

  5. #5 peak

    every year so there are already lots of instances where these characters have been created.⏎By drawing inspiration from those, you would not only be paying homage to their work, but also improving your hand by practicing, exploring, and creating new ways to convey

  6. #6 peak

    , fostering customer relationships, and meeting business goals instead of constantly verifying the integrity of their data.⏎When data is dirty, it not only affects the productivity of teams, but it can also potentially hurt relationships with customers or damage the reputation of the business

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: rather than rather than rather than rather than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than than

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

    employable or that your career prospects are good enough to allow you to pursue a career path of your dreams.⏎Different industry sectors demand different employability skills – personal as well as technical. Most employer surveys do, however, note that personal skills have

  2. #2 peak

    the world. Moreover, the pattern of suicides in China is quite different than in other parts of the world – there are more completed suicides among females than males and rural rates are three-fold urban rates. The lack of reliable suicide data prior to

  3. #3 peak

    is the sequence of events in the conversion of the chemical energy derived from ATP hydrolysis to a mechanical stroke? How are these altered in a processive motor, where each of the catalytic heads is subject to mechanical forces from intramolecular

  4. #4 peak

    who are more experienced, adults as well as inexperienced citizens such as children, not only in the knowledge of values but

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. L15:3061 strongly suppresses
  2. L15:2926 strongly excites
  3. L14:1830 excites
  4. L15:1922 excites
  5. L16:533 suppresses
  6. L15:1055 suppresses
  7. L14:2500 suppresses
  8. L14:2386 excites
  9. L15:454 excites
  10. L14:864 suppresses
  11. L13:2418 excites
  12. L14:2535 suppresses
  13. L15:2448 suppresses
  14. L12:2918 excites
  15. L16:1084 suppresses
  16. L13:20 suppresses
  17. L12:537 excites
  18. L14:779 suppresses 'than' in comparisons
  19. L15:2542 suppresses
  20. L15:791 excites
  21. L13:737 excites
  22. L15:2640 excites
  23. L12:1987 excites
  24. L15:2015 excites

Feeds later layers

  1. L22:1337 strongly suppresses
  2. L19:2514 strongly suppresses
  3. L19:2308 strongly suppresses
  4. L21:1090 suppresses
  5. L21:2310 suppresses
  6. L20:1309 suppresses
  7. L18:2844 suppresses
  8. L18:3059 suppresses
  9. L27:1088 suppresses
  10. L18:1553 suppresses
  11. L27:2851 suppresses
  12. L21:478 suppresses
  13. L18:2331 suppresses
  14. L27:568 suppresses
  15. L18:2007 suppresses
  16. L18:2761 suppresses
  17. L19:2976 suppresses
  18. L22:1248 suppresses
  19. L21:783 suppresses
  20. L21:1597 suppresses
  21. L21:1008 suppresses
  22. L23:1127 suppresses
  23. L23:1955 suppresses
  24. L20:1646 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 -6.13 and -5.791 values between -5.79 and -5.461 values between -5.13 and -4.801 values between -4.47 and -4.141 values between -3.80 and -3.473 values between -3.47 and -3.144 values between -2.81 and -2.485 values between -2.48 and -2.144 values between -2.14 and -1.818 values between -1.81 and -1.4816 values between -1.48 and -1.1531 values between -1.15 and -0.82107 values between -0.82 and -0.49477 values between -0.49 and -0.152726 values between -0.15 and 0.18372 values between 0.18 and 0.5161 values between 0.51 and 0.8434 values between 0.84 and 1.1726 values between 1.17 and 1.515 values between 1.51 and 1.845 values between 1.84 and 2.175 values between 2.17 and 2.501 values between 2.50 and 2.831 values between 2.83 and 3.163 values between 3.16 and 3.502 values between 3.50 and 3.832 values between 3.83 and 4.161 values between 4.16 and 4.491 values between 4.49 and 4.821 values between 5.15 and 5.491 values between 5.82 and 6.151 values between 6.81 and 7.15offnegativepositiveoften

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. ionales
  2. ioni
  3. ện
  4. 개인정보
  5. IÓN
  6. 丘
  7. _mobile
  8. commerce
  9. onden
  10. очные

Suppresses

  1. versus
  2. vs
  3. 而不是
  4. 而非
  5. rather
  6. instead
  7. rather
  8. instead
  9. ではなく
  10. 而是