Neuron Explorer

Qwen3-0.6B

opening bracket after an acronym's full name

A neuron in layer 24 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 reacts to the opening bracket that comes right after a long organisation or term name, just before its short form, like 'Health Maintenance Organizations (HMOs)'. Its negative side fires on letter labels in lists, which seems to be a different pattern. Turning it up or down gave text with no clear link to acronyms.

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

    , a number of crashes involving teen drivers have led to more than a dozen teenagers tragically losing their lives. And a recent report by the Governors Highway Safety Association (GHSA) shows that teen driver fatalities are on the rise among 16

  2. #2 peak

    Many Americans receive medical insurance coverage through their employers. These plans are typically a combination of employer and employee contributions and offer a range of options, including Health Maintenance Organizations (HMOs), Preferred Provider Organizations (PPOs), and High-Deduct

  3. #3 peak

    hydrogen fuel cell car fleet may hit top gear within five years, but only if there is enough investment in filling stations, the UK Hydrogen and Fuel Cells Association (UK HFCA) told Reuters on Friday.⏎Fuel cells convert hydrogen into electricity,

  4. #4 peak

    venous leg ulcers cite medical compression as the standard treatment for chronic venous conditions.⏎In the open ulcer stage, the Haute Autorité de Santé (HAS) [French National Authority for Health

  5. #5 peak

    describes three different methodologies which could be applied to distribute the burden and the benefits of a global carbon budget in a fair and equitable way.⏎- Greenhouse development rights (GDRs), where<|endoftext|>Hi, Can anyone please help me in clearing the

  6. #6 peak

    (Andolu Agency, 2020; Pulse Nigeria, 2020). The World Health Organization has labeled Nigeria a human resources for health (HRH) crisis country (Global Health Workforce Alliance, 202

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: The government of the United States is made of people who have personal information that they might have access to. This personal information might be accessed and personal information might be stored and manipulated, and also personal information might be manipulated and manipulated beyond the

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

    . Solution : X = 2 Z e) Find two solutions of A X = Y 1 . Solution : X = V and X = V + Z f) Find another solution of A X = 3 Y 1- 5

  2. #2 peak

    ) Alfred Nobel - Dynamite⏎(c) Alexander Graham Bell - Telephone⏎(d) Michael Faraday - Dynamo⏎(e) William Sholes - Typewriter⏎(f) Louis Daguerre - Photography⏎- (a) Lewis Waterman⏎

  3. #3 peak

    4) Associate Justice d) беспартийная система⏎5) a right of voting e) гражданин государства⏎6) district court f) гражданские и уголовные дела⏎7) a resident of the state

  4. #4 peak

    Telephone⏎(b) Alfred Nobel Dynamo⏎(c) Alexander Graham Bell Thermos Flask⏎(d) Michael Faraday Dynamite⏎(e) William Sholes Photography⏎(f) Louis Daguerre Typewriter⏎- Name the inventors of the following

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. L22:73 strongly suppresses
  2. L23:1042 strongly suppresses
  3. L22:59 excites
  4. L23:1941 excites
  5. L20:2992 suppresses
  6. L21:2435 suppresses
  7. L23:3026 suppresses
  8. L22:2696 excites
  9. L23:956 excites
  10. L18:46 suppresses
  11. L22:2884 excites
  12. L20:3017 weakly suppresses
  13. L21:1605 weakly suppresses
  14. L16:2944 weakly suppresses
  15. L15:2928 weakly excites
  16. L18:34 weakly excites
  17. L23:150 weakly suppresses
  18. L22:1091 weakly excites
  19. L16:2902 weakly suppresses
  20. L22:1567 weakly suppresses
  21. L19:907 weakly excites
  22. L22:742 weakly suppresses
  23. L3:392 weakly suppresses
  24. L23:1385 weakly excites

Feeds later layers

  1. L27:524 strongly suppresses
  2. L27:1155 excites just before a pronoun for God
  3. L25:573 excites
  4. L27:1071 excites
  5. L26:1664 weakly excites
  6. L27:874 weakly suppresses
  7. L26:1147 weakly suppresses
  8. L27:1090 weakly excites
  9. L26:1310 weakly suppresses
  10. L27:859 weakly suppresses unclear: token just before a k sound
  11. L26:297 weakly suppresses
  12. L27:272 weakly excites
  13. L25:726 weakly excites
  14. L25:1989 weakly excites
  15. L26:2631 weakly suppresses
  16. L27:219 weakly excites
  17. L25:1961 weakly suppresses
  18. L25:496 weakly suppresses
  19. L26:630 weakly excites
  20. L27:93 weakly excites
  21. L25:961 weakly excites years and numbers before the digits
  22. L26:2966 weakly suppresses
  23. L25:333 weakly suppresses
  24. L27:2658 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 -13.09 and -12.351 values between -8.66 and -7.931 values between -7.19 and -6.452 values between -6.45 and -5.711 values between -4.98 and -4.241 values between -4.24 and -3.503 values between -3.50 and -2.7616 values between -2.76 and -2.0328 values between -2.03 and -1.29132 values between -1.29 and -0.552721 values between -0.55 and 0.19913 values between 0.19 and 0.9251 values between 0.92 and 1.6619 values between 1.66 and 2.402 values between 2.40 and 3.142 values between 3.14 and 3.875 values between 3.87 and 4.615 values between 4.61 and 5.351 values between 6.08 and 6.821 values between 6.82 and 7.561 values between 7.56 and 8.301 values between 15.67 and 16.41offnegativepositiveoften

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. 祂
  2. His
  3. _G
  4. He
  5. (G
  6. Him
  7. G
  8. G
  9. GH
  10. _H

Suppresses

  1. h
  2. h
  3. h
  4. =h
  5. hoff
  6. (h
  7. g
  8. "g
  9. |h
  10. *h