Neuron Explorer

Qwen3-0.6B

slot right before a quantity

A neuron in layer 8 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 becomes active on the empty spot just before a number, such as "the other 47 states" or "losing just 5 to 10 percent." It seems to expect a count or amount to come next. Its negative side appears on words like "fifth" and "percent." Steering changed the topic toward maths and numbers, but not consistently.

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

    6 percent). Only one state, New Hampshire, had a poverty rate of less than ten percent for women, at 8.9 percent. In the other 47 states and the District of Columbia, between 10 and 2

  2. #2 peak

    60 second-, third-, and fourth-grade students, while 48 second- and fourth-grade students were subjects in the other 3 studies. Results indicated that: (1) younger children tended to have difficulty

  3. #3 peak

    warming has already reduced rainfall and increased evaporation in the Mideast, worsening water shortages. Up to now, climate scientists had projected that rainfall could decline another 20 percent by 2100. But the Dead Sea cores suggest

  4. #4 peak

    ics, for another 75,000 to 1.75 million years. After

  5. #5 peak

    at increased risk for type 2 diabetes, heart disease, osteoarthritis, and certain forms of cancer.⏎âIf youâre overweight, losing just 5 to 10 percent of your body weight can slash the risk of both

  6. #6 peak

    counting 1, 2, 3, 4 like this:⏎The first note is held for 4 counts, the second bar each note is held for 2 counts and, in the

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 impact of a natural disaster on the results of the calculation. So, what are the effects of a natural disaster on the results of the calculation? I have been trying to find a way to find the effect of a natural disaster on

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

    greatest challenges.⏎Roughly 1.7 million children under five years old die every year in India, says Unicef, representing more than a fifth of all child deaths worldwide. Most of these deaths are preventable as they stem

  2. #2 peak

    , the Greeting Card Association notes that over 1 billion valentines are sent annually, vs. 2.5 billion Christmas cards; 85 percent of all valentines are purchased by women.⏎Vintage Valentines are increasingly popular

  3. #3 peak

    —makes up less than one-half of 1 percent of what’s out there. Most of the universe is unseen

  4. #4 peak

    are unsafe within the confines of the military system. According to the Department of Defense, almost a quarter of all women in the military experience sexual harassment, and more than 20,000 soldiers were sexually assaulted in the 20

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. L7:1742 suppresses
  2. L7:1524 suppresses
  3. L6:1469 weakly excites
  4. L1:1095 weakly excites
  5. L6:2006 weakly excites
  6. L7:224 weakly suppresses
  7. L1:1802 weakly suppresses
  8. L3:2177 weakly suppresses
  9. L7:1388 weakly excites
  10. L6:1562 weakly excites
  11. L7:1114 weakly suppresses
  12. L7:583 weakly suppresses
  13. L6:457 weakly excites
  14. L5:276 weakly suppresses
  15. L4:1555 weakly suppresses
  16. L6:345 weakly excites
  17. L5:2724 weakly suppresses
  18. L7:531 weakly suppresses
  19. L1:1847 weakly suppresses
  20. L3:293 weakly excites
  21. L5:2029 weakly excites
  22. L7:2496 weakly excites
  23. L1:2281 weakly suppresses
  24. L7:398 weakly excites

Feeds later layers

  1. L17:222 excites
  2. L17:1570 excites
  3. L9:97 suppresses
  4. L16:1897 excites
  5. L11:579 excites
  6. L14:471 excites
  7. L12:1467 excites numbers after dates and figures
  8. L9:335 suppresses 'of' after nouns like anniversary or onset
  9. L10:2063 excites
  10. L10:868 excites
  11. L16:399 excites
  12. L12:408 suppresses
  13. L25:1212 excites
  14. L13:2915 excites
  15. L13:855 excites
  16. L18:1259 excites
  17. L19:1419 excites
  18. L13:1308 excites
  19. L15:2091 excites
  20. L13:122 suppresses
  21. L21:676 excites
  22. L25:46 suppresses
  23. L11:751 excites numbers and quantities mid-phrase
  24. L19:1117 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 -4.98 and -4.761 values between -2.85 and -2.632 values between -2.42 and -2.213 values between -2.21 and -1.992 values between -1.78 and -1.577 values between -1.57 and -1.357 values between -1.35 and -1.1414 values between -1.14 and -0.9312 values between -0.93 and -0.7132 values between -0.71 and -0.5068 values between -0.50 and -0.29609 values between -0.29 and -0.072604 values between -0.07 and 0.14457 values between 0.14 and 0.3557 values between 0.35 and 0.5712 values between 0.57 and 0.785 values between 0.78 and 0.991 values between 0.99 and 1.213 values between 1.21 and 1.424 values between 1.42 and 1.631 values between 1.85 and 2.062 values between 2.06 and 2.271 values between 2.27 and 2.491 values between 3.12 and 3.342 values between 3.34 and 3.55offnegativepositiveoften

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. 三千
  2. nero
  3. 五千
  4. eron
  5. significant
  6. 几百
  7. 几千
  8. 众多
  9. banyak
  10. semiclass

Suppresses

  1. jec
  2. äng
  3. etten
  4. sujet
  5. нашего
  6. uck
  7. ận
  8. ться
  9. uve
  10. agle