Neuron Explorer

Qwen3-0.6B

'this' after 'earlier' or 'in'

A neuron in layer 6 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 fires on the word 'this' in phrases about time or place, such as 'earlier this year', 'in this area' or 'this month'. Turning it up did not give a clear, related change in the replies.

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

    colleges and universities and public school systems have embarked on aggressive efforts to overhaul science curriculums, laboratories and teacher training. President Clinton's education legislation passed earlier this year by Congress, the Goals 2000: Educate America Act

  2. #2 peak

    Zed Seselja described the UN criticisms as “bizarre”.⏎The UN report followed harsh interim and final reports on the state of affairs in Australia earlier this year by the UN’s Special Rapporteur on the Rights of Indigenous People Victoria

  3. #3 peak

    simple premise had yet to be tested. John enlisted coleopterist Peter Kovarik as a partner in this study. The two decided that in addition to butterflies, scarabaeine scar

  4. #4 peak

    clamation project continues to clean up the polluted canal.⏎The Malting factory has been abandoned since 1989 when Canadian National stopped rail service to industries in this area of the canal. The clay silos are protected as part of the L

  5. #5 peak

    be a watershed moment in the fight to stop climate change, as Congress returns from its recess to take up the energy bill and consider landmark global warming legislation. Earlier this year

  6. #6 peak

    of these occurred among adults between the ages of 25 and 34, a number high enough to mean that 20% of all deaths in this age group in 2016 involved opioids.⏎Among those between the ages

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 number of the days on the wall of a refrigerator, and the number of the bottles in the blue boxes, and the number of the bottles in the red boxes, and the number of the bottles in the green boxes, and the number

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

    “Look now toward heaven, and count the stars if you are able to number them…. So

  2. #2 peak

    giveth of his bread to the poor.” – Proverbs 22:9⏎2 Corinthians 9:6-8 – “But this I say, He which soweth sparingly shall reap also sparingly; and he which

  3. #3 peak

    The African American Migration Experience, the Schomburg Center for Research in Black Culture.⏎- Chicago and the Great Migration, The Newberry.⏎- "Sir I Will Thank You with All My Heart": Seven Letters from the Great Migration, History Matters

  4. #4 peak

    proclaimed, "Here we will build a temple to our God." When the city was surveyed, the block enclosing

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. L5:98 suppresses
  2. L4:722 excites
  3. L5:546 excites
  4. L4:1762 excites "this" in time and context phrases
  5. L5:730 excites
  6. L4:1940 suppresses
  7. L5:108 suppresses
  8. L5:520 weakly excites
  9. L2:793 weakly excites
  10. L5:2242 weakly excites
  11. L2:996 weakly excites
  12. L5:1779 weakly suppresses
  13. L3:1135 weakly excites
  14. L5:1000 weakly excites
  15. L3:889 weakly suppresses
  16. L4:0 weakly suppresses
  17. L5:552 weakly excites the word 'below' pointing to content further down
  18. L5:965 weakly suppresses
  19. L5:2968 weakly excites
  20. L5:1282 weakly excites
  21. L5:1869 weakly excites
  22. L4:1855 weakly excites
  23. L3:1677 weakly excites
  24. L4:992 weakly excites

Feeds later layers

  1. L23:2652 weakly excites
  2. L21:333 weakly suppresses
  3. L16:116 weakly suppresses
  4. L17:2255 weakly excites
  5. L11:658 weakly excites
  6. L9:314 weakly suppresses
  7. L24:1283 weakly excites
  8. L13:1654 weakly suppresses
  9. L17:1482 weakly excites
  10. L26:2063 weakly suppresses
  11. L14:236 weakly suppresses 'and' joining two parallel subjects
  12. L21:270 weakly suppresses
  13. L14:1938 weakly excites
  14. L20:163 weakly suppresses
  15. L15:357 weakly suppresses
  16. L17:408 weakly suppresses
  17. L22:1353 weakly excites
  18. L25:2147 weakly excites 'World' in organisation names
  19. L16:166 weakly suppresses
  20. L10:741 weakly suppresses
  21. L12:2235 weakly excites
  22. L10:1941 weakly suppresses
  23. L26:1718 weakly excites
  24. L21:124 weakly 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 -2.83 and -2.681 values between -2.53 and -2.382 values between -1.64 and -1.492 values between -1.49 and -1.345 values between -1.19 and -1.043 values between -1.04 and -0.894 values between -0.89 and -0.746 values between -0.74 and -0.5915 values between -0.59 and -0.4464 values between -0.44 and -0.29457 values between -0.29 and -0.141948 values between -0.14 and 0.011140 values between 0.01 and 0.16166 values between 0.16 and 0.3046 values between 0.30 and 0.4511 values between 0.45 and 0.606 values between 0.60 and 0.756 values between 0.75 and 0.904 values between 0.90 and 1.055 values between 1.05 and 1.209 values between 1.20 and 1.351 values between 1.35 and 1.501 values between 1.50 and 1.651 values between 1.65 and 1.802 values between 1.80 and 1.951 values between 2.84 and 2.991 values between 2.99 and 3.14offnegativepositiveoften

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. ried
  2. rí
  3. thon
  4. ">--}}⏎
  5. ley
  6. isson
  7. ária
  8. rozen
  9. oundation
  10. aign

Suppresses

  1. изма
  2. 有条件的
  3. tempt
  4. 固定的
  5. 不说
  6. 感冒
  7. 说着
  8. 冷漠
  9. 重复
  10. enti