Neuron Explorer

Qwen3-0.6B

copula after 'probably' or 'may'

A neuron in layer 15 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

It tends to fire on the verb right after a hedge word like 'probably', 'may' or 'is', when the sentence is judging whether something is true or false, as in myths and warnings. Turning it up made the model repeat questions and phrases, so the effect is unclear.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    than 88cm (34 inches)⏎..... HDL cholesterol less<|endoftext|>Geese have the reputation of attacking innocent bystanders like walkers and golfers. Geese become aggressive because they don’t like to share their space with humans and

  2. #2 peak

    far out, get a cramp, suffer an injury, or get caught in a current.⏎Myth: Dry or secondary drowning is not a real thing.⏎Reality: While rare (and these terms are considered medically obsolete), breathing problems can

  3. #3 peak

    like the one on the right. Its open end can point up, down, left or right. A curve of this shape is called 'parabolic', meaning 'like a parabola'.⏎Quick navigation: Geometric Figure⏎: A u

  4. #4 peak

    before you fill out any forms, or give out money or credit card details.⏎- Don't accept any offers that seem too good to be true - they probably are.⏎Remember to always be

  5. #5 peak

    ata' signifies. It is truly said that whatever is in the world is in the Mahabharata, and what is not in Mahabharata is not in the world; thus stating a simple fact that answers to issues that we face

  6. #6 peak

    for you, don’t worry. Enjoy the ease of formatting equations in TeX, along with all the benefits of a What You See Is What You Get (WYSIWYG).⏎What if you could simply insert your equations into your document with

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: I can't be sure that I can't be... I can't... I can... I can I... I... I... I... I... I I... I I... I I I... I I I... I I I I

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

    in the soil, beneficial soil bacteria ate the calcium citrate and excreted the calcium in a form that allowed it to react quickly with sodium phosphate to form calcium apatite, which covered the sand and soil particles underground, forming the sponge

  2. #2 peak

    Prevention. But thankfully, blood pressure can usually be lowered and managed through simply making better lifestyle choices.⏎And just as lowering your cholesterol naturally is better than using statins, lowering your blood pressure is best done without relying solely on medication. Let’s

  3. #3 peak

    myth, Tefnut became angry with her father Re and ran off to Nubia. The god Thoth went to find her and conjoled her into returning to Egypt.⏎

  4. #4 peak

    were excellent patients. They were pretty healthy, with a mean body mass index of 26 — just barely overweight — and only 3.6 percent of them smoked. They were also good sports: By the end of the trial, more

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. L14:1593 suppresses
  2. L12:202 excites
  3. L14:124 suppresses
  4. L11:111 excites
  5. L13:627 suppresses
  6. L13:230 excites
  7. L12:1843 suppresses
  8. L8:10 suppresses
  9. L13:385 excites
  10. L14:101 suppresses
  11. L10:2410 suppresses
  12. L13:290 suppresses
  13. L11:1134 excites
  14. L11:1391 excites
  15. L10:1452 suppresses
  16. L11:1377 suppresses
  17. L13:255 weakly excites
  18. L13:489 weakly excites
  19. L12:583 weakly suppresses
  20. L14:494 weakly suppresses
  21. L14:65 weakly suppresses
  22. L13:310 weakly suppresses
  23. L10:490 weakly suppresses
  24. L6:13 weakly suppresses

Feeds later layers

  1. L16:1346 excites
  2. L16:329 suppresses
  3. L16:1090 suppresses
  4. L16:1149 suppresses
  5. L20:176 excites
  6. L16:1359 suppresses
  7. L16:2209 suppresses
  8. L16:631 excites
  9. L16:510 excites
  10. L18:1536 suppresses end of web page disclaimer
  11. L16:1861 suppresses
  12. L16:392 suppresses
  13. L16:2921 excites
  14. L19:839 suppresses
  15. L17:198 excites
  16. L16:1903 suppresses
  17. L18:997 excites
  18. L16:972 weakly suppresses
  19. L16:1743 weakly excites
  20. L18:497 weakly suppresses
  21. L18:199 weakly excites
  22. L19:821 weakly excites
  23. L16:966 weakly suppresses
  24. L18:157 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 -4.50 and -4.322 values between -2.72 and -2.543 values between -2.54 and -2.377 values between -2.37 and -2.199 values between -2.19 and -2.019 values between -2.01 and -1.8316 values between -1.83 and -1.6511 values between -1.65 and -1.4838 values between -1.48 and -1.3055 values between -1.30 and -1.1294 values between -1.12 and -0.94120 values between -0.94 and -0.77212 values between -0.77 and -0.59274 values between -0.59 and -0.41395 values between -0.41 and -0.23449 values between -0.23 and -0.05552 values between -0.05 and 0.13434 values between 0.13 and 0.30373 values between 0.30 and 0.48280 values between 0.48 and 0.66192 values between 0.66 and 0.84124 values between 0.84 and 1.0190 values between 1.01 and 1.1968 values between 1.19 and 1.3739 values between 1.37 and 1.5518 values between 1.55 and 1.7318 values between 1.73 and 1.907 values between 1.90 and 2.0811 values between 2.08 and 2.266 values between 2.26 and 2.441 values between 2.44 and 2.62offnegativepositiveoften

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. Invoker
  2. :async
  3. 렌
  4. 辞
  5. レン
  6. język
  7. oggles
  8. iska
  9. aprend
  10. ренд

Suppresses

  1. xAA
  2. orse
  3. xBA
  4. Pais
  5. Phoenix
  6. xbe
  7. Cooper
  8. PECT
  9. Napoleon
  10. xBB