Neuron Explorer

Qwen3-0.6B

'as' in 'as low as' and 'as far as'

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 word 'as' in set phrases such as 'as low as', 'as far as' and 'as early as'. Turning it up made the model produce long strings of the word 'as'.

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

    McKinley or Denali (its Koyukon Athabaskan name meaning “The Great One”). The majestic mountain, which has extremely cold weather with temperatures as low as -75.5 °F (−60 °C)

  2. #2 peak

    shade the window lightly with a sheer curtain.⏎Keep the room temperature between 65 and 80 degrees Fahrenheit. Although tropical houseplants can survive in temperatures as low as 55 F, growth will be stunted. Avoid putting tropical

  3. #3 peak

    company’s DNA testing kit you use which determines how far back a DNA test will go. Another factor is what type of test you use, as some can go as far back as your original origins, though there are restrictions on what you can find

  4. #4 peak

    markets and many nonprofit organizations have coalesced in support of open source software. Even governments have heralded their support of open source software and some have even gone as far as to initiate mandates that render the distribution of open source software all but compulsory

  5. #5 peak

    existence and origin of life in terms of mere coincidences. the roots of this philosophy go back as far as antiquity and ancient Greece.⏎All atheist philosophies that deny creation,

  6. #6 peak

    the device cools down, making it self-regenerating. Most importantly, their approach works at temperatures as low as 30 degrees Celsius.⏎“Other researchers have studied electrochemical heat

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: zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip zip

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

    5 4 3 2 1⏎We need to use the range function to iterate over the array using a for loop. The range() function returns a sequence of numbers starts from a given number to another given number with a given step

  2. #2 peak

    is not necessary to know it,⏎but in some cases the difference between shared and separated objects yields different orders of space or time complexity.⏎Consider the infinite list⏎let x =

  3. #3 peak

    or not and take the default route. Here’s an example of⏎mwhen for Monoids:⏎-- Important numbers!⏎xs = [6, 9, 13, 42]⏎-- Without

  4. #4 peak

    to reduce values, e.g. lists have the very popular⏎foldr. This package provides that same function for the⏎Bool data type; a function called⏎bool function is a replacement for the build-in⏎if then else-syntax. However

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. L20:1918 suppresses opening bracket after a unit conversion
  2. L19:1222 suppresses
  3. L17:1859 suppresses
  4. L8:1430 weakly excites
  5. L19:2701 weakly suppresses
  6. L18:569 weakly excites
  7. L1:290 weakly suppresses
  8. L1:2080 weakly excites
  9. L15:2797 weakly suppresses
  10. L21:1797 weakly suppresses
  11. L4:750 weakly excites
  12. L11:2008 weakly suppresses
  13. L20:2184 weakly excites
  14. L20:2834 weakly suppresses
  15. L2:1825 weakly excites
  16. L15:865 weakly excites
  17. L14:459 weakly excites
  18. L15:1139 weakly suppresses
  19. L12:1709 weakly excites
  20. L16:833 weakly excites
  21. L0:2060 weakly excites
  22. L1:2096 weakly excites
  23. L9:1815 weakly excites
  24. L7:2094 weakly excites

Feeds later layers

  1. L26:588 excites
  2. L26:2729 suppresses
  3. L27:849 suppresses
  4. L27:471 excites
  5. L25:1157 excites
  6. L27:679 suppresses
  7. L25:2236 suppresses
  8. L23:390 weakly excites
  9. L24:1904 weakly suppresses
  10. L25:2003 weakly suppresses
  11. L23:1246 weakly excites
  12. L24:797 weakly suppresses
  13. L23:1422 weakly excites
  14. L27:1862 weakly suppresses
  15. L23:710 weakly suppresses
  16. L25:2641 weakly suppresses music and art genre words
  17. L26:2494 weakly excites
  18. L25:620 weakly excites
  19. L24:2939 weakly suppresses
  20. L25:2240 weakly suppresses
  21. L24:734 weakly suppresses
  22. L23:2179 weakly suppresses
  23. L24:2380 weakly suppresses
  24. L26:1281 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 -3.82 and -3.191 values between -2.56 and -1.9313 values between -1.93 and -1.29190 values between -1.29 and -0.662090 values between -0.66 and -0.031545 values between -0.03 and 0.6033 values between 0.60 and 1.239 values between 1.23 and 1.868 values between 1.86 and 2.496 values between 2.49 and 3.122 values between 3.12 and 3.751 values between 3.75 and 4.393 values between 4.39 and 5.021 values between 5.02 and 5.651 values between 7.54 and 8.171 values between 15.11 and 15.741 values between 17.01 and 17.641 values between 20.16 and 20.791 values between 20.79 and 21.42offnegativepositiveoften

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. as
  2. 作为
  3. far
  4. As
  5. As
  6. →as
  7. -as
  8. AS
  9. as
  10. 作為

Suppresses

  1. zip
  2. (zip
  3. zip
  4. Zip
  5. Zip
  6. .zip
  7. ZIP
  8. enumerate
  9. ZIP
  10. _zip