Neuron Explorer

Qwen3-0.6B

'something' after is or not

A neuron in layer 7 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 word 'something' when it follows a phrase like 'is', 'is not', or 'was'. Turning it up did not visibly change what the model wrote, so this is only a word pattern.

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

    may impact your brain—even in moderation.⏎Drinking in excess—or binge drinking—is something that most everyone agrees can speed up your brain's aging process. But what is

  2. #2 peak

    pandemic? It’s certainly not something we want to experience again anytime soon, or ever.⏎I know San Diego,

  3. #3 peak

    The Christmas holiday would also be a time that some slaveowners gave wine and alcoholic beverages to their slaves. With business still in mind, the effects of alcohol were something unknown to many slaves

  4. #4 peak

    .E.<|endoftext|>Danny HarrisHonors English 1 Smoking and TobaccoSmoking and Tobacco in general is one of the major issues in our country. Smoking is not something people should be doing. It is costly, not only, because of the financial

  5. #5 peak

    tartar buildup in all or most of the teeth. Your cat is likely around 3 to 5 years old if there is. Wear-and-tear is something you still have to look out for, and this is usually observed when kittens are

  6. #6 peak

    than cutting them down, that it plans to encourage residents to treat their trees by extending to them the rates the city receives for pesticide injection. “Treating is something we promote as a city,” Council Member Dan Kealey said when the city adopted

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: "A, b" is a function of a variable, and "c" is a function of a variable, and so on, and the rest is a function of a variable, and so on, and the rest is a function of

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

    -ray technologies and advanced computational techniques will provide a clear and detailed picture of the human brain and allow more effective treatments of brain diseases.⏎Singapore, 15 January 2020 Scientists from Taiwan initiated an international coalition that will produce the first

  2. #2 peak

    your home if left untreated. A dehumidifier or air conditioner can help.⏎Answer: True. Radon—a colorless, tasteless and odorless gas—seeps into homes through cracks in walls, floors and foundations. If rad

  3. #3 peak

    6 is the most common. It is an alpha-emitter with 1601 years of half-life.⏎Sources of radon⏎Radon is a gas produced by the decay of radium. An estimated 1 gram of radium

  4. #4 peak

    little girl with a thin neck, and spindling legs half-hidden in the folds of Aunt Frances’s skirts. But she didn’t<|endoftext|>Dec 6, 2023⏎Airbus has successfully passed the Preliminary Design Review (

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. L0:2283 excites
  2. L4:139 excites
  3. L3:1111 suppresses
  4. L3:76 suppresses
  5. L6:2237 excites
  6. L5:1139 suppresses
  7. L4:118 weakly excites
  8. L4:24 weakly excites
  9. L2:1414 weakly suppresses
  10. L0:918 weakly excites
  11. L2:949 weakly suppresses
  12. L3:409 weakly excites
  13. L1:878 weakly suppresses
  14. L5:1317 weakly suppresses 'a' right after 'being'
  15. L1:2509 weakly excites
  16. L0:359 weakly suppresses hyphen inside COVID-19 and similar compounds
  17. L5:378 weakly suppresses
  18. L4:44 weakly suppresses
  19. L0:2739 weakly excites
  20. L5:395 weakly excites
  21. L1:2975 weakly suppresses
  22. L2:49 weakly suppresses
  23. L5:1876 weakly excites
  24. L2:866 weakly excites

Feeds later layers

  1. L11:507 strongly excites
  2. L13:2727 excites
  3. L15:738 excites
  4. L15:1668 excites
  5. L10:235 excites
  6. L8:386 excites
  7. L18:1415 excites
  8. L10:647 excites
  9. L16:1715 excites
  10. L13:2835 excites
  11. L17:2756 excites
  12. L15:668 excites
  13. L15:309 suppresses
  14. L17:510 excites
  15. L16:2038 excites
  16. L10:417 excites
  17. L10:1382 excites
  18. L10:2575 excites
  19. L12:738 excites
  20. L18:2949 excites
  21. L17:390 excites
  22. L16:2926 excites
  23. L21:586 weakly suppresses
  24. L9:152 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 -2.66 and -2.561 values between -2.26 and -2.161 values between -2.06 and -1.961 values between -1.47 and -1.371 values between -1.17 and -1.071 values between -1.07 and -0.971 values between -0.97 and -0.872 values between -0.87 and -0.773 values between -0.77 and -0.675 values between -0.67 and -0.578 values between -0.57 and -0.4729 values between -0.47 and -0.3787 values between -0.37 and -0.28272 values between -0.28 and -0.18598 values between -0.18 and -0.081280 values between -0.08 and 0.021018 values between 0.02 and 0.12435 values between 0.12 and 0.22116 values between 0.22 and 0.3226 values between 0.32 and 0.429 values between 0.42 and 0.523 values between 0.52 and 0.623 values between 0.62 and 0.721 values between 0.72 and 0.823 values between 0.82 and 0.921 values between 0.92 and 1.022 values between 1.22 and 1.31offnegativepositiveoften

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. ensive
  2. -negative
  3. 一样的
  4. .NONE
  5. 지를
  6. anten
  7. Clement
  8. 好的
  9. 性的
  10. 她说

Suppresses

  1. //{{
  2. ớt
  3. erties
  4. ptions
  5. Portions
  6. арат
  7. /company
  8. rol
  9. леж
  10. 笆