Neuron Explorer

Qwen3-0.6B

'as' introducing a reason

A neuron in layer 20 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

Fires on the word 'as' when it starts a short explanation of why something is true, like 'do not use detergent as a substitute'. Turning it down made the model repeat 'as', and turning it up made it repeat 'it's'.

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

    If this happens, add a watering agent such as Water-In to enhance soil penetration. Water-In is available at most garden stores. Do not use detergent as a substitute, as this will be toxic to the plant.⏎Heavy clay soils drain

  2. #2 peak

    or as a takeout container from a fast food place⏎- egg cartons – either the styrofoam ones or cardboard ones (latter is preferable as you can plant these in the ground and they will rot away)⏎- plastic containers

  3. #3 peak

    ulent is indoors, it’s a good idea to get a small spout watering can. It’s also a good idea to avoid using high pressure streams of water as this can damage the roots of your succulent.⏎Another way to gauge the moisture

  4. #4 peak

    of lego bricks⏎If you have completed this algorithm, the final answer should be approximatively:⏎It is good practice to use subroutines whenever possible as these can then easily be reused in other programs. Your task is to adapt your

  5. #5 peak

    the tar utility are tagged with .tgz or .tar.gz. Tar creates an archive for your files by combining multiple files. This is a widely used format as it saves a lot of space and bandwidth while installing downloading and installing the packages.⏎

  6. #6 peak

    and small leaves.⏎Actually, goats are recommended to consume more “green things” than fruits as it provides them with enough nutrients that facilitate the digestive process. Then, strawberry tops

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: as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as as

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

    data, only 2 out of 10 households meet the estimated average requirement (EAR) of Vitamin A in their daily diet. This trend is particularly worrying, given the rising rates of food insecurity that arose as a consequence of the COVID-

  2. #2 peak

    with emerging technologies and frameworks is crucial to deliver cutting-edge apps that cater to users’ evolving expectations. Moreover, debugging and troubleshooting are constant companions throughout the development process, demanding problem-solving skills and meticulous attention to detail from developers.⏎Key Factors for Successful

  3. #3 peak

    keep the crew alive for several months would be cut down tremendously, as sustenance could be delivered intravenously by robots. Exercise equipment would no longer be necessary, as the robots could stimulate major muscle groups periodically. Overall, the ships would be

  4. #4 peak

    using technology and other tools, individuals can work smarter, not harder, and achieve their goals in less time. Additionally, smart work can lead to innovation and creativity, as individuals are encouraged to think outside the box and find new solutions to old problems

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. L19:2440 suppresses
  2. L19:1036 suppresses
  3. L19:2296 suppresses
  4. L18:1841 suppresses
  5. L18:892 suppresses
  6. L19:2321 excites
  7. L16:1112 suppresses
  8. L19:975 suppresses
  9. L17:2364 excites
  10. L18:1003 excites
  11. L17:1311 excites
  12. L19:91 suppresses
  13. L19:1380 excites
  14. L17:2154 suppresses
  15. L18:442 weakly excites
  16. L15:2038 weakly suppresses
  17. L17:637 weakly suppresses comma opening an interrupting phrase
  18. L13:948 weakly excites
  19. L6:250 weakly excites
  20. L17:580 weakly excites
  21. L17:2462 weakly excites
  22. L6:2638 weakly excites
  23. L19:1084 weakly excites "and" joining clauses after a condition
  24. L16:2236 weakly excites

Feeds later layers

  1. L21:191 suppresses
  2. L23:298 excites
  3. L22:689 excites comma closing an intro clause
  4. L21:639 excites
  5. L26:588 suppresses
  6. L27:717 excites
  7. L23:839 suppresses comma before a trailing 'with' clause
  8. L21:655 excites
  9. L24:123 suppresses
  10. L21:1445 excites
  11. L27:1244 excites
  12. L21:434 excites
  13. L27:471 suppresses
  14. L22:892 excites
  15. L27:607 excites
  16. L21:172 excites
  17. L21:513 excites empty slot before a decade
  18. L27:1095 excites
  19. L22:1805 weakly excites
  20. L22:2261 weakly suppresses
  21. L21:2045 weakly excites unclear: weak noun and pronoun activity
  22. L22:353 weakly suppresses
  23. L25:880 weakly suppresses comma after a cause or effect clause
  24. L23:147 weakly excites "that" introducing a clause or "that is"

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 -15.51 and -14.841 values between -14.84 and -14.171 values between -13.50 and -12.832 values between -12.83 and -12.161 values between -10.81 and -10.143 values between -10.14 and -9.472 values between -9.47 and -8.801 values between -8.80 and -8.131 values between -8.13 and -7.462 values between -7.46 and -6.791 values between -6.79 and -6.123 values between -5.45 and -4.786 values between -4.78 and -4.118 values between -4.11 and -3.4411 values between -3.44 and -2.7713 values between -2.77 and -2.1020 values between -2.10 and -1.43104 values between -1.43 and -0.761177 values between -0.76 and -0.092299 values between -0.09 and 0.58152 values between 0.58 and 1.2627 values between 1.26 and 1.9316 values between 1.93 and 2.6014 values between 2.60 and 3.2711 values between 3.27 and 3.9410 values between 3.94 and 4.614 values between 4.61 and 5.2810 values between 5.28 and 5.951 values between 5.95 and 6.623 values between 6.62 and 7.292 values between 7.29 and 7.961 values between 10.64 and 11.31offnegativepositiveoften

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. 它可以
  2. 它
  3. they
  4. it
  5. there
  6. 它们
  7. it
  8. they
  9. 它是
  10. 这些

Suppresses

  1. as
  2. →as
  3. 为您
  4. as
  5. with
  6. /as
  7. As
  8. 作为
  9. 列为
  10. (as