Neuron Explorer

Qwen3-0.6B

number after 'account for' or 'over'

A neuron in layer 22 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 spot just before a statistic or large number, such as 'account for 63%' or 'over 13 million'. Turning it up made the model talk about statistics and data, and turning it down did too.

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

    1):155⏎Despite a century of frequently effective preventative and control initiatives, infectious illnesses continue to be a major worldwide public health issue, resulting in over 13 million annual fatalities. The rise of novel illnesses, the resurgence of diseases

  2. #2 peak

    diseases and diabetes, are responsible for 63% of all deaths worldwide (36 million out 57 million

  3. #3 peak

    for people and society.<|endoftext|>Fact 1: The world’s population is rapidly ageing⏎The number of people aged 60 years or older will rise from 900 million to 2 billion between 2015 and

  4. #4 peak

    a concise, action-oriented outcome document that will shape the global agendas for generations to come.⏎10 Facts On NCDS⏎1. NCDs account for 63% of all deaths.⏎Noncommunicable diseases (NCDs),

  5. #5 peak

    are trending the wrong way.⏎Noncommunicable diseases – or NCDs – like heart attacks and strokes, cancers, diabetes and chronic respiratory disease account for over 63% of deaths

  6. #6 peak

    an infection. It develops when the chemicals the immune system releases into the bloodstream to fight an infection cause inflammation throughout the entire body instead.⏎The illness causes an estimated 14 million deaths every year.⏎In a new study by the University of British

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: statistics and performance data analytics - you can't do better than the overall performance statistics you aggregate - you can't reach higher performance statistics - you can't reach higher performance - you can't reach higher performance - you can't reach higher performance -

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

    PL). The figure was 15.2 per cent in 2006. This shows a commendable reduction of poverty. The Department of Census and Statistics (DCS) declares that the OPL is Rs. 3,

  2. #2 peak

    more<|endoftext|>Growth Trends for Related Jobs⏎Physicians and surgeons are some of the highest-paid and most well-educated individuals in the country. According to the Bureau of Labor Statistics, there were 661,400 individuals employed as

  3. #3 peak

    temporary employment agencies earned some of the lowest salaries, an average of $46,360 a year.⏎According to the Bureau of Labor Statistics, electricians working in the West and Northeast regions of the United States

  4. #4 peak

    weekly earnings by worker characteristics, including educational attainment on the right-hand side:⏎Source: New York Times, Bureau of Labor Statistics⏎For Rampell, the figure makes a slam-dunk case for

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. L10:2772 suppresses
  2. L9:2953 excites
  3. L21:1545 excites
  4. L8:2625 excites
  5. L21:1941 suppresses
  6. L10:503 excites
  7. L20:1966 excites
  8. L19:1224 suppresses
  9. L21:317 excites
  10. L9:1136 suppresses
  11. L6:2851 suppresses
  12. L21:1607 suppresses
  13. L20:1678 suppresses
  14. L20:2545 suppresses
  15. L18:1885 suppresses
  16. L6:2129 suppresses
  17. L11:991 excites
  18. L11:917 excites
  19. L9:333 excites
  20. L6:922 suppresses
  21. L19:922 excites
  22. L21:686 excites
  23. L18:2352 weakly excites
  24. L11:1921 weakly suppresses

Feeds later layers

  1. L27:2715 strongly suppresses
  2. L23:803 suppresses
  3. L27:698 suppresses
  4. L23:2207 suppresses
  5. L25:1863 suppresses
  6. L26:1107 suppresses
  7. L26:961 suppresses
  8. L23:2120 suppresses
  9. L27:1568 suppresses
  10. L23:919 suppresses
  11. L26:181 suppresses
  12. L24:1151 suppresses
  13. L25:853 suppresses
  14. L23:1965 suppresses
  15. L23:1667 suppresses
  16. L24:1795 suppresses
  17. L25:2657 suppresses
  18. L25:1486 suppresses
  19. L24:2680 suppresses
  20. L24:2758 suppresses
  21. L26:2778 suppresses
  22. L26:292 weakly suppresses
  23. L23:1063 weakly suppresses
  24. L25:1691 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 -22.42 and -21.491 values between -17.78 and -16.851 values between -14.06 and -13.141 values between -12.21 and -11.281 values between -11.28 and -10.352 values between -10.35 and -9.421 values between -9.42 and -8.493 values between -8.49 and -7.561 values between -7.56 and -6.636 values between -6.63 and -5.7111 values between -5.71 and -4.7810 values between -4.78 and -3.8525 values between -3.85 and -2.9234 values between -2.92 and -1.9977 values between -1.99 and -1.06439 values between -1.06 and -0.132872 values between -0.13 and 0.80377 values between 0.80 and 1.7311 values between 1.73 and 2.659 values between 2.65 and 3.586 values between 3.58 and 4.514 values between 4.51 and 5.447 values between 5.44 and 6.372 values between 6.37 and 7.301 values between 7.30 and 8.233 values between 8.23 and 9.151 values between 11.94 and 12.871 values between 13.80 and 14.73offnegativepositiveoften

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. osition
  2. kos
  3. uri
  4. 熹
  5. spoilers
  6. ично
  7. Arms
  8. inge
  9. itch
  10. .Getenv

Suppresses

  1. 统计
  2. 统计数据
  3. statistics
  4. statistical
  5. stats
  6. Statistics
  7. statistics
  8. Statistical
  9. statist
  10. statistic