Neuron Explorer

Qwen3-0.6B

'about' or 'approximately' before a number

A neuron in layer 24 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 lights up on the words 'about' and 'approximately' when they come right before a number or amount, like 'about 27%' or 'about 3,000'. Turning it down made the model repeat 'about' and 'approximately' over and over.

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

    drop off from 7 million jobs in 2006 to 4.5 million in 2009.⏎Part of the decline is that about 27% fewer businesses with employees are being started, the study says.⏎

  2. #2 peak

    that in 2015 the number of mobile phone subscribers in North Korea reached 3.24 million, (about 13%) and that about 60 percent of the population in Pyongyang, the capital, between 2

  3. #3 peak

    unit of GDP should go down by 40% - 45% from the 2005 level, non-fossil energy should make up about 15% of primary energy consumption, and forest coverage should increase by

  4. #4 peak

    for U.S. teens, accounting for more than one in three deaths, according to the Centers for Disease Control and Prevention. In 2009, about 3,000 15–to-19-year-olds

  5. #5 peak

    is a leading cause of poor quality of life and annually there’s an estimated 9M people falling ill with and 1.5M dying of it. About 4000 people die of TB every day in the whole world,

  6. #6 peak

    2015. The opioid crisis is largely responsible for the increase—for example, two thirds of overdose deaths are associated with opioid misuse. Surprisingly, about half of all opioid overdose deaths are related to medications from prescriptions (National Academy of

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: approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately approximately

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

    6 trillion degrees Celsius (10.8 trillion °F)! Definitely hotter than the temperature of the centre of the Sun, which comes in at a puny 5 million °C (9 billion °F).⏎This record temperature

  2. #2 peak

    being used.⏎Here`s a blurb about their current design process :⏎Primary concept and development work takes place at the Billund⏎headquarters, where the company employs approximately 120 designers.⏎The company also has smaller design offices in the

  3. #3 peak

    these melanomas occur in preexisting moles.⏎How quickly can a mole turn cancerous?⏎Melanoma can grow very quickly. It can become life-threatening in as little as 6 weeks and, if untreated, it can spread to other

  4. #4 peak

    astronomy . Accreted matter is usually hydrogen rich. It forms the outer envelope of a neutron star, which contains a hydrogen burning shell, with an energy release of about 5 MeV/nucleon in a stable burning. Helium ashes

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. L20:780 strongly excites 'of' before a measured value
  2. L23:1514 strongly suppresses
  3. L18:291 strongly suppresses
  4. L23:947 strongly excites
  5. L19:1921 strongly excites
  6. L23:2791 strongly suppresses
  7. L20:2316 strongly suppresses
  8. L22:481 strongly excites
  9. L19:1222 strongly suppresses
  10. L23:1723 strongly suppresses
  11. L21:175 strongly suppresses
  12. L22:2408 strongly excites
  13. L21:326 suppresses
  14. L23:1919 suppresses
  15. L10:868 suppresses
  16. L20:1001 suppresses
  17. L19:1211 excites
  18. L16:1919 suppresses
  19. L20:1998 suppresses
  20. L19:2151 excites
  21. L21:1284 excites
  22. L13:2125 suppresses
  23. L15:2797 suppresses
  24. L22:685 suppresses

Feeds later layers

  1. L27:1058 strongly suppresses
  2. L27:1705 strongly suppresses
  3. L25:857 suppresses
  4. L25:1212 suppresses
  5. L25:2963 suppresses
  6. L25:2908 suppresses
  7. L25:1264 suppresses
  8. L25:734 suppresses
  9. L25:1136 suppresses
  10. L25:2543 suppresses
  11. L26:2770 suppresses
  12. L25:1369 suppresses
  13. L25:2902 suppresses
  14. L25:1381 suppresses
  15. L26:2537 suppresses
  16. L25:992 weakly suppresses
  17. L25:921 weakly suppresses
  18. L25:1020 weakly suppresses
  19. L25:2035 weakly suppresses
  20. L25:329 weakly suppresses
  21. L27:1595 weakly suppresses
  22. L27:628 weakly suppresses
  23. L26:474 weakly suppresses
  24. L26:1848 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 -34.94 and -33.281 values between -29.97 and -28.321 values between -20.04 and -18.392 values between -18.39 and -16.733 values between -16.73 and -15.087 values between -15.08 and -13.424 values between -13.42 and -11.778 values between -11.77 and -10.1116 values between -10.11 and -8.4618 values between -8.46 and -6.8023 values between -6.80 and -5.1560 values between -5.15 and -3.49121 values between -3.49 and -1.84354 values between -1.84 and -0.183222 values between -0.18 and 1.4748 values between 1.47 and 3.131 values between 3.13 and 4.784 values between 4.78 and 6.443 values between 6.44 and 8.101 values between 8.10 and 9.751 values between 9.75 and 11.412 values between 11.41 and 13.062 values between 13.06 and 14.721 values between 19.68 and 21.341 values between 26.30 and 27.962 values between 27.96 and 29.611 values between 29.61 and 31.27offnegativepositiveoften

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. 态
  2. onomous
  3. pez
  4. ingham
  5. ndl
  6. isms
  7. 责任制
  8. Single
  9. uite
  10. ismo

Suppresses

  1. 约
  2. 約
  3. approximately
  4. 约为
  5. about
  6. 大约
  7. approximately
  8. about
  9. around
  10. About