Neuron Explorer

Qwen3-0.6B

unclear: mid-sentence words before a noun

A neuron in layer 18 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

This neuron fires on a range of ordinary words in the middle of sentences, often right before the main noun of a phrase, such as 'chemical', 'dairy' or 'neurological'. The pattern is loose and hard to pin down. Turning it up made the model's writing wander into unrelated topics.

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

    Chemical Activity of Copper:⏎The chemical activity of copper imparts unique taste and aromatic characteristics to beverages.⏎Despite their higher

  2. #2 peak

    bodies, and pigment granules. Understanding these components will not only deepen your knowledge of hair but also help you identify them and comprehend their significance. So, let's dive in!⏎The cuticle is the outermost layer of the hair shaft,

  3. #3 peak

    atty Liver Disease?⏎Non-alcoholic fatty liver disease typically does not present notable symptoms. To be clear, alcoholic fatty liver disease has few early warning signs, too. Symptoms

  4. #4 peak

    for someone like me<|endoftext|>By Sarah Zagorski⏎In the United Kingdom, women in their early thirties are now more likely to have abortions than teenagers, according to a

  5. #5 peak

    pace. To switch between modelspace and paperspace, view the tabs located at the bottom of the screen. One tab is labeled ‘modelspace’ and the other tabs are labeled either as ‘sheet’

  6. #6 peak

    Tourette syndrome is a neurological condition that involves automatic, unconscious movements and utterances called tics. Tourette

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: what if we can find a way to control the control of the control of the control of the control of the control of the control of the control of the control of the control of the control of the control of the control of the control 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

    paintings.⏎Sextus Empiricus' Outlines of Pyrrhonism is among the most crucial and influential texts within the historical past of Greek philosophy. within the calls for of cause Casey Perin

  2. #2 peak

    & Social Sciences Communications prompt youngsters skilled psychological well being issues associated to collaborating in lively shooter drills. Within the giant research, a group of statisticians from the Georgia Institute of Know-how

  3. #3 peak

    use of the phrases hope, love, dwelling, college, youngsters, group, help, and assist after the drills. The researchers thought of posting with these phrases within the aftermath of the drills to be indicative of getting excessive nervousness.⏎They included

  4. #4 peak

    floor group again on Earth helps the astronauts. Flight controllers comply with a process plan that is round 30 pages lengthy, however there are different plans in place ought

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. L17:581 suppresses
  2. L10:557 excites
  3. L16:704 excites
  4. L15:2567 suppresses
  5. L17:19 suppresses
  6. L9:1378 excites
  7. L11:308 excites
  8. L10:196 suppresses
  9. L15:6 excites
  10. L10:1255 excites
  11. L10:1050 weakly suppresses end of document marker token
  12. L14:84 weakly excites
  13. L3:112 weakly excites
  14. L7:983 weakly excites
  15. L17:27 weakly suppresses
  16. L15:337 weakly suppresses
  17. L15:910 weakly suppresses
  18. L16:16 weakly excites
  19. L9:260 weakly suppresses
  20. L14:2465 weakly suppresses
  21. L10:1523 weakly suppresses
  22. L1:2160 weakly suppresses
  23. L2:2246 weakly suppresses
  24. L14:2937 weakly excites

Feeds later layers

  1. L19:901 strongly excites
  2. L24:2694 strongly excites
  3. L19:2035 strongly excites
  4. L22:599 strongly excites
  5. L20:748 strongly excites
  6. L27:1096 strongly excites
  7. L23:733 strongly excites
  8. L27:1299 strongly excites
  9. L26:2773 strongly excites
  10. L19:1193 excites
  11. L22:1317 excites unclear: sentence and clause breaks in essays
  12. L19:440 suppresses
  13. L26:2727 excites
  14. L25:2469 excites
  15. L21:229 excites
  16. L21:677 excites
  17. L19:1303 excites
  18. L21:3012 excites
  19. L22:795 excites
  20. L26:2715 excites
  21. L22:238 excites
  22. L24:602 excites
  23. L20:1277 suppresses
  24. L25:171 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 -1.18 and -1.121 values between -1.12 and -1.061 values between -1.06 and -1.003 values between -0.94 and -0.883 values between -0.88 and -0.8212 values between -0.82 and -0.7711 values between -0.77 and -0.7133 values between -0.71 and -0.6548 values between -0.65 and -0.5986 values between -0.59 and -0.53140 values between -0.53 and -0.47195 values between -0.47 and -0.41307 values between -0.41 and -0.35353 values between -0.35 and -0.30398 values between -0.30 and -0.24462 values between -0.24 and -0.18456 values between -0.18 and -0.12379 values between -0.12 and -0.06380 values between -0.06 and -0.00256 values between -0.00 and 0.06135 values between 0.06 and 0.1279 values between 0.12 and 0.1859 values between 0.18 and 0.2332 values between 0.23 and 0.2915 values between 0.29 and 0.3519 values between 0.35 and 0.4113 values between 0.41 and 0.474 values between 0.47 and 0.536 values between 0.53 and 0.593 values between 0.59 and 0.652 values between 0.65 and 0.703 values between 0.70 and 0.763 values between 0.76 and 0.823 values between 0.82 and 0.882 values between 0.88 and 0.942 values between 0.94 and 1.001 values between 1.00 and 1.062 values between 1.12 and 1.17offnegativepositiveoften

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. 黉
  2. employing
  3. 国度
  4. swiftly
  5. accom
  6. populace
  7. employs
  8. 若
  9. na
  10. uela

Suppresses

  1. )'),⏎
  2. **)
  3. 吃过
  4. JOR
  5. '))->
  6. '])->
  7. UserController
  8. ** ⏎
  9. -----------*/⏎
  10. VERR