Neuron Explorer

Qwen3-0.6B

unclear: the word 'kind' and mixed words

A neuron in layer 22 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on a grab bag of words, most often 'kind' (as in 'a kind gesture' or 'kind of funny'), plus pieces of 'Creative' and a few others. Turning it up or down did not produce any clear change in what the model wrote.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    a part of it. Open and hold doors for people. Let someone else go before you in a grocery store line just because it is a nice gesture. Act kind in front

  2. #2 peak

    electrical resistance,⏎- I is the electric current,⏎- V is the voltage.⏎(Note: this is not true where the impedance is complex)⏎|This page uses Creative Commons Licensed content from Wikipedia (view authors).|<|endoftext|>Of the moon that is

  3. #3 peak

    part as an image that "the deep heart of nature" (perhaps God) may look more kind on the errant Hester and her child than her Puritan neighbors do.⏎

  4. #4 peak

    Success in Cirrcuit lies⏎Too bright for our infirm Delight⏎The Truth's superb surprise⏎As Lightening to the Children eased⏎With explanation kind⏎The Truth must dazzle gradually⏎Or every man be blind---⏎1.

  5. #5 peak

    impulsively and irrationally. Salekin uses a laptop computer and LCD player to explain brain plasticity

  6. #6 peak

    uba diver trained to work in underwater installations and conduct, assist in, or be a subject of scientific research. Also called oceanaut.⏎from Wiktionary, Creative Commons Attribution/Share-Alike License⏎- n. An underwater explorer.⏎from

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: there is a 24 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

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

    protector deity Penden Lhamo⏎- The Goddess of the White Umbrella (the Buddhist deity Ushnisha-sitatapatra)⏎- The Great mystic Virupa⏎- Thunderbolt and bell⏎- Prayer wheel⏎- Cabinet

  2. #2 peak

    old<|endoftext|>Immune cells in the brain have surprising influence on sexual behavior⏎Relatively uncommon mast cells appear powerful in development, study finds⏎Researchers have found a surprising new explanation

  3. #3 peak

    actual glimpse of its mystic history. From ancient buildings to the modern wonders Delhi offers anything one can desire

  4. #4 peak

    circle. Man today is not wiser than the ancient seers and sages and thinkers, not more spiritual than the great seekers of old, the first mighty mystics, not superior in arts and crafts to the ancient artists and craftsmen;

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. L19:2402 weakly suppresses
  2. L19:107 weakly excites
  3. L20:601 weakly suppresses
  4. L17:148 weakly excites
  5. L19:2924 weakly suppresses
  6. L20:570 weakly excites
  7. L21:237 weakly suppresses
  8. L18:420 weakly suppresses
  9. L6:374 weakly excites
  10. L9:1213 weakly excites
  11. L14:1187 weakly excites
  12. L20:2446 weakly suppresses
  13. L19:277 weakly suppresses
  14. L11:2297 weakly suppresses
  15. L20:380 weakly excites
  16. L5:2738 weakly suppresses
  17. L21:2189 weakly suppresses
  18. L3:2866 weakly suppresses
  19. L12:819 weakly excites
  20. L0:2370 weakly suppresses
  21. L13:971 weakly excites
  22. L16:2063 weakly suppresses
  23. L11:865 weakly excites
  24. L21:16 weakly suppresses

Feeds later layers

  1. L23:520 weakly suppresses
  2. L27:2044 weakly suppresses
  3. L23:2965 weakly suppresses
  4. L24:2833 weakly suppresses
  5. L23:1882 weakly suppresses
  6. L24:561 weakly suppresses
  7. L23:2034 weakly excites
  8. L25:2661 weakly suppresses
  9. L27:1178 weakly excites
  10. L27:2810 weakly suppresses
  11. L27:908 weakly suppresses
  12. L23:2070 weakly excites
  13. L24:2463 weakly suppresses
  14. L24:2050 weakly suppresses
  15. L24:800 weakly suppresses colon in clock times
  16. L23:1210 weakly excites
  17. L25:1556 weakly excites
  18. L23:1831 weakly suppresses
  19. L27:1316 weakly suppresses
  20. L25:618 weakly suppresses
  21. L26:649 weakly suppresses
  22. L26:1994 weakly suppresses
  23. L24:1251 weakly suppresses
  24. L25:787 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 -25.78 and -24.612 values between -21.08 and -19.901 values between -19.90 and -18.731 values between -15.20 and -14.022 values between -14.02 and -12.852 values between -11.67 and -10.502 values between -10.50 and -9.323 values between -9.32 and -8.154 values between -8.15 and -6.979 values between -6.97 and -5.7918 values between -5.79 and -4.6232 values between -4.62 and -3.4449 values between -3.44 and -2.27113 values between -2.27 and -1.092156 values between -1.09 and 0.091357 values between 0.09 and 1.2683 values between 1.26 and 2.4422 values between 2.44 and 3.6118 values between 3.61 and 4.7912 values between 4.79 and 5.978 values between 5.97 and 7.144 values between 7.14 and 8.324 values between 8.32 and 9.492 values between 9.49 and 10.671 values between 10.67 and 11.841 values between 11.84 and 13.021 values between 20.07 and 21.25offnegativepositiveoften

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. eri
  2. eria
  3. rists
  4. adam
  5. alty
  6. ng
  7. Side
  8. emy
  9. edy
  10. Side

Suppresses

  1. agnitude
  2. ounding
  3. iqué
  4. icine
  5. 做一个
  6. ulaire
  7. obox
  8. uild
  9. less
  10. iciel