Neuron Explorer

Qwen3-0.6B

verb slot after a long subject

A neuron in layer 26 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the small word that comes right after a long subject and before the main verb, such as 'have', 'has' or 'likely' in a sentence. Turning it up made the model repeat words like 'provided' and 'caused', which fits the kind of verb that would come next.

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

    ing the states of Indiana, Ohio, Illinois, and Michigan. Convenient access to raw materials, skilled labor, and the shipping options available including the railroads, likely contributed to the concentration of these companies within this area. Another factor was a receptive

  2. #2 peak

    cultural contexts—from Appalachia and the rural Piedmont to New Orleans, the Black Belt of the Deep South, and the Carolina Lowcountry. These uniquely Southern settings have spawned equally unique musical forms.⏎Visit Event Page HERE!<|endoftext|>NORD is very

  3. #3 peak

    atic and European tectonic plates. The melting of glaciers and rise in sea levels after the last Ice Age, 20,000 years ago, made the mountains into an archipelago which was then eroded by the sea

  4. #4 peak

    team who are providing guidance to a literacy program for the displaced Massalit people of Darfur. Her upbringing, coupled with experience living in a Western culture, has become an

  5. #5 peak

    Supreme Court—failed the movement for black freedom and representation, the amendment process was critical to shaping the fight for racial equality. Indeed, it was the amendment process that created the legal basis to abolish slavery and assure equal political rights for black Americans.

  6. #6 peak

    I transformed humanitarian relief. The magnitude of the conflict, the extent of civilian suffering, particularly in occupied areas, and the destruction and consequences after the end of hostilities accelerated the rationalization of humanitarian relief. Organizations, both temporary and permanent, embraced

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: how can continued satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames satenames sat

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

    148 species assessed for population and distribution trends, 60 per cent have declined over the past 50 years; further research shows turtle dove numbers have plummeted by 80 per cent since 1995, hedge

  2. #2 peak

    ⏎Although the number of operating longwall mining systems in the U.S. has remained relatively constant over the past five years, U.S. longwall production levels have increased significantly during this period. Longwalls account for over 40% of

  3. #3 peak

    that the number of new employer businesses has fallen 27% since 2006. The total number of startups (employer businesses and self-employed) has increased since the recession, according to the Kauffman Index of Entrepreneurial Activity.

  4. #4 peak

    simply did not exist. Prices fell through the 1980s but have slowly rebounded through the 90s, and in recent years prices have tended to fluctuate between $3 and $10 per board-foot, or

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. L24:60 strongly excites
  2. L23:535 excites
  3. L23:127 suppresses
  4. L22:444 suppresses
  5. L19:2475 excites
  6. L22:479 excites
  7. L21:958 suppresses
  8. L20:2672 excites
  9. L24:239 suppresses
  10. L22:2258 excites
  11. L18:178 suppresses
  12. L21:1181 excites
  13. L23:108 suppresses
  14. L20:349 excites
  15. L25:848 suppresses
  16. L16:1532 excites
  17. L19:2161 excites
  18. L7:92 suppresses
  19. L24:778 excites
  20. L20:2524 suppresses
  21. L15:390 suppresses
  22. L15:184 suppresses
  23. L21:773 excites
  24. L19:1678 suppresses

Feeds later layers

  1. L27:1514 excites
  2. L27:1295 excites
  3. L27:1596 excites
  4. L27:2836 weakly excites
  5. L27:1567 weakly suppresses
  6. L27:2757 weakly excites
  7. L27:217 weakly excites
  8. L27:869 weakly excites
  9. L27:1275 weakly suppresses
  10. L27:820 weakly suppresses
  11. L27:988 weakly excites word before 'needed' or 'required'
  12. L27:1345 weakly excites
  13. L27:410 weakly suppresses
  14. L27:1946 weakly excites the letter z inside words
  15. L27:1876 weakly excites
  16. L27:382 weakly suppresses
  17. L27:1372 weakly suppresses
  18. L27:3068 weakly suppresses
  19. L27:2524 weakly excites
  20. L27:727 weakly suppresses
  21. L27:3070 weakly suppresses
  22. L27:1619 weakly suppresses
  23. L27:716 weakly excites
  24. L27:2703 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 -30.30 and -28.453 values between -24.75 and -22.904 values between -22.90 and -21.052 values between -21.05 and -19.201 values between -19.20 and -17.346 values between -17.34 and -15.492 values between -15.49 and -13.648 values between -13.64 and -11.7913 values between -11.79 and -9.9428 values between -9.94 and -8.0946 values between -8.09 and -6.24106 values between -6.24 and -4.39177 values between -4.39 and -2.54598 values between -2.54 and -0.692554 values between -0.69 and 1.16176 values between 1.16 and 3.0157 values between 3.01 and 4.8631 values between 4.86 and 6.7119 values between 6.71 and 8.5613 values between 8.56 and 10.4111 values between 10.41 and 12.2613 values between 12.26 and 14.118 values between 14.11 and 15.966 values between 15.96 and 17.816 values between 17.81 and 19.663 values between 19.66 and 21.513 values between 21.51 and 23.362 values between 23.36 and 25.222 values between 25.22 and 27.071 values between 27.07 and 28.922 values between 28.92 and 30.771 values between 30.77 and 32.622 values between 36.32 and 38.172 values between 38.17 and 40.021 values between 41.87 and 43.72offnegativepositiveoften

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. provided
  2. caused
  3. Provided
  4. produced
  5. created
  6. provided
  7. generated
  8. offered
  9. supplied
  10. made

Suppresses

  1. ς
  2. уть
  3. 下了
  4. enames
  5. 笑了
  6. 让孩子
  7. usz
  8. _Metadata
  9. 맥
  10. sat