Neuron Explorer

Qwen3-0.6B

'of' after a percentage

A neuron in layer 14 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 'of' when it comes right after a percentage or a share, as in '40% of families'. Turning it up pushed the model toward number and percentage talk, such as repeated figures and ratios.

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

    the Johns Hopkins University and sponsored by the National Institutes of Health.⏎In the study, the most common allergy was to peanuts, with about 1.5 percent of people testing highly positive for the peanut antibodies, tiny proteins made by the body when

  2. #2 peak

    Whites who lived in such neighborhoods have fared much better: Only 40% of those families still live in the poorest neighborhoods.⏎Making matters worse, the value of

  3. #3 peak

    13 and 28 are now ten times more likely to live in poor neighborhoods than their white counterparts. The same study found that 67% of Black families who, a generation ago, lived in the poorest 25%

  4. #4 peak

    survey of more than 9,000 people, asked why people with mental health and substance abuse disorders did not seek treatment. About 15% of the respondents<|endoftext|>Okay, I am by no means knowledgable in this field

  5. #5 peak

    of diagnosing individuals thought to have contracted HIV.⏎A prospective study of HIV infection and the development of AIDS in subjects with hemophilia showed that a much larger proportion of HIV-infected persons would develop AIDS than previously thought (Goedert et al

  6. #6 peak

    nea include older age, male sex, and higher BMI. In a new study of 1448 people living with Parkinson's, 45% of participants experienced sleep apnea. In addition to the general risk factors, for people

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: I have found that I have been thinking of these ideas, and that this is the first time, the previous ones, or even the first. This is a new and different ideas, and this is the first, or the previous ones,

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

    , each on the national, state, and district levels. In addition, we generated state maps with color-coded congressional districts showing the average annual change in employment from 2011 to 2020. Each district within a state

  2. #2 peak

    cancer when compared with women who did less than 3 MET hours of physical activity in a week. The rate of survival from breast cancer related death increases with increasing physical activities done per week and is the maximum

  3. #3 peak

    the premise of family/whanau, social connectedness, culture and community. Social services has at its heart a focus on social justice and an equitable access to resources for all. The latter may include access to adequate housing, health and educational

  4. #4 peak

    including lipsticks, foundations and mascaras, and found some of the products to contain significant levels of PFAS.⏎In the initial tests, researchers found high levels of fluorine in most waterproof mascaras and liquid lipsticks. Fluorine can

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. L13:844 excites 'of' after a percentage or share
  2. L13:712 excites
  3. L13:1794 suppresses
  4. L13:554 suppresses 'of' linking a noun to a larger whole
  5. L13:1535 suppresses
  6. L13:2636 suppresses
  7. L13:787 suppresses
  8. L12:1738 excites
  9. L13:291 suppresses
  10. L13:1518 excites
  11. L12:850 excites
  12. L13:1082 suppresses
  13. L11:846 weakly suppresses
  14. L11:1040 weakly excites
  15. L12:1653 weakly suppresses
  16. L12:329 weakly excites
  17. L6:1731 weakly suppresses
  18. L12:248 weakly suppresses
  19. L3:27 weakly suppresses
  20. L13:2440 weakly excites
  21. L13:2375 weakly excites
  22. L11:666 weakly excites
  23. L10:2142 weakly excites
  24. L12:2347 weakly suppresses

Feeds later layers

  1. L16:805 strongly excites
  2. L15:958 excites
  3. L16:619 suppresses
  4. L17:865 excites
  5. L15:1028 excites
  6. L16:1934 excites
  7. L15:428 suppresses
  8. L16:2987 suppresses
  9. L15:748 excites
  10. L17:2749 excites
  11. L16:1096 excites
  12. L16:1831 excites
  13. L15:216 suppresses
  14. L16:2581 excites
  15. L18:457 excites
  16. L15:414 suppresses
  17. L16:1984 excites
  18. L15:146 suppresses
  19. L18:1701 excites
  20. L20:1615 excites
  21. L15:349 excites
  22. L18:2349 excites
  23. L17:1221 excites
  24. L16:680 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 -2.98 and -2.801 values between -2.08 and -1.901 values between -1.90 and -1.722 values between -1.72 and -1.535 values between -1.35 and -1.176 values between -1.17 and -0.995 values between -0.99 and -0.8113 values between -0.81 and -0.6346 values between -0.63 and -0.45162 values between -0.45 and -0.27547 values between -0.27 and -0.091906 values between -0.09 and 0.10886 values between 0.10 and 0.28241 values between 0.28 and 0.4648 values between 0.46 and 0.6415 values between 0.64 and 0.825 values between 0.82 and 1.005 values between 1.00 and 1.183 values between 1.18 and 1.363 values between 1.36 and 1.552 values between 1.55 and 1.734 values between 1.73 and 1.911 values between 4.08 and 4.26offnegativepositiveoften

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. ��
  2. Player
  3. 诀
  4. ソー
  5. $args
  6. quotes
  7. piel
  8. stitial
  9. asonry
  10. ность

Suppresses

  1. ])).
  2. 磬
  3. .Unique
  4. fbe
  5. rang
  6. 最少
  7. 给她
  8. ione
  9. .have
  10. }`).