Neuron Explorer

Qwen3-0.6B

'a' right after 'played' or 'play'

A neuron in layer 27 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 fires on the word 'a' in phrases like 'played a key role' or 'plays a crucial part', right before the word that describes the role. Turning it down made the writing repeat 'a' and 'other' in a loop, but turning it up gave text that did not clearly stay on the topic of roles. Only the trigger is clear here.

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

    really interested in the fundamental mechanisms that drive brain development and sex-specific brain development, and this study found that mast cells – immune cells involved in allergic responses – play a key role,” Lenz said.⏎If human development mirrors what was seen in this

  2. #2 peak

    While addressing modifiable factors such as diet and exercise can have a deep positive impact on our health and lower high cholesterol levels, it turns out that genetics also plays a crucial part in how high our cholesterol may end up

  3. #3 peak

    illa Stuckey-Kauffman⏎During the early nineteenth century American Protestants sprang into missionary action, sending workers to countries around the world. Women played a

  4. #4 peak

    800 new street trees per year that will be maintained by the Department of Transportation.⏎The city’s publicly owned utility, Seattle City Light, has also played a significant role in expanding and maintaining Seattle’s urban forest. Through its urban tree replacement

  5. #5 peak

    to another with the purpose of sending a particular meaning or idea represented by the meme. The spread of memes has significantly shaped how people communicate, and they have played a fundamental role in advancing the contemporary digital culture. Memes provide a compelling new way

  6. #6 peak

    to treat bipolar disorder have similar functions and effects, a psychiatrist might prescribe a variety of medications as part of the client’s overall treatment plan. Mental health professionals play a pivotal role in consistently monitoring and ensuring that the recommended psychopharmacological treatments are

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: art att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att att

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

    the digits are⏎consecutive. How many times like this are there between midnight⏎and 7 a.m.?⏎These practical challenges are all about making a 'tray' and covering it with paper.⏎My cousin was 24 years

  2. #2 peak

    ⏎implement that pseudocode that do not require the last check, but none of real⏎"I have come to believe that the whole world is an enigma, a harmless enigma⏎that is made terrible by our own mad attempt to interpret it

  3. #3 peak

    at IBM in Switzerland that had developed a complex technique to image or “take a picture” of molecules in the journal Nature — it turned out to be the coup de grace for the team.⏎Back in 2009, at the IBM

  4. #4 peak

    a matrix⏎Find nth maximum⏎Matlab Basics II - Create a vector with a repeated entry⏎Create sequnce 1 4 9 16

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. L23:703 strongly suppresses
  2. L26:2810 strongly suppresses 'a' before 'lot', 'little' or 'bit'
  3. L19:751 strongly excites the article 'a' before a noun
  4. L26:190 strongly suppresses
  5. L23:266 strongly suppresses
  6. L22:83 strongly suppresses
  7. L22:1881 excites
  8. L17:188 excites 'a' before a noun phrase
  9. L25:366 suppresses
  10. L18:3 excites
  11. L16:880 suppresses
  12. L19:3056 excites
  13. L20:901 suppresses
  14. L4:137 excites
  15. L26:562 suppresses unclear: list items and sentence breaks
  16. L19:733 excites
  17. L21:1967 excites
  18. L22:2384 excites
  19. L9:59 suppresses "a" meaning per (a day, a year)
  20. L24:296 suppresses
  21. L16:2693 suppresses
  22. L10:3 suppresses
  23. L22:441 excites
  24. L4:31 weakly excites

Feeds later layers

None: this is the last layer.

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.

407 values between -6.76 and -3.013383 values between -3.01 and 0.7435 values between 0.74 and 4.481 values between 8.23 and 11.981 values between 11.98 and 15.732 values between 19.47 and 23.221 values between 26.97 and 30.712 values between 34.46 and 38.211 values between 38.21 and 41.951 values between 45.70 and 49.451 values between 49.45 and 53.202 values between 68.18 and 71.932 values between 71.93 and 75.685 values between 75.68 and 79.433 values between 79.43 and 83.175 values between 83.17 and 86.925 values between 86.92 and 90.673 values between 90.67 and 94.416 values between 94.41 and 98.166 values between 98.16 and 101.917 values between 101.91 and 105.655 values between 105.65 and 109.406 values between 109.40 and 113.154 values between 113.15 and 116.904 values between 116.90 and 120.641 values between 120.64 and 124.394 values between 124.39 and 128.141 values between 128.14 and 131.881 values between 131.88 and 135.633 values between 139.38 and 143.13offnegativepositiveoften

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. utors
  2. 那时候
  3. nodiscard
  4. woke
  5. ffects
  6. venues
  7. .k
  8. uras
  9. 路上
  10. unts

Suppresses

  1. -other
  2. att
  3. other
  4. existing
  5. array
  6. attachment
  7. extension
  8. additional
  9. OTHER
  10. atom