Neuron Explorer

Qwen3-0.6B

'the' before a noun mid-sentence

A neuron in layer 16 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 responds to the word 'the' in the middle of a sentence, especially after a preposition or verb, where a noun phrase is about to start, as in 'close to the skin' or 'away from the previous point'. It turns negative on 'the' in phrases like 'players of the official game'. Turning it up produced repetitive filler and loops rather than clear meaning.

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

    :⏎Teaching Strategies and Tips⏎Use this lesson to show how exponential functions and geometric sequences are related.⏎- In a geometric sequence, terms are found by multiplying the same constant to the previous term.⏎- For exponential functions, when the input variable

  2. #2 peak

    this by multiplying the number by the maximum possible mark. Once you’ve determined the maximum, you can also divide it by the total marks. This will give you the percentage of a given number. In addition to finding the actual value of a number

  3. #3 peak

    function that is part of the standard library (and uses the length of the data as the seed). It is quite insecure in a cryptographic sense, but since an educated cr

  4. #4 peak

    Web page: http://www.wider.unu.edu/⏎More information through EDIRC⏎Please report citation or reference errors to , or , if you are the registered author of the cited work, log in to your RePEc Author Service

  5. #5 peak

    visit our Prevention page. Briefly though:⏎- Use fine-point tweezers or a special tick-removing tool. Grasp the tick as close to the skin as possible. If you don’t have tweezers, protect your fingers

  6. #6 peak

    2.5 and the learning rate is 0.01, then the gradient descent algorithm will pick the next point 0.025 away from the previous point.⏎If you pick a learning rate that is too small, learning will

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 need to be more of a "dualist" in the way of the old days of the "soul of the people" of the firsts. (if we are not going to get the "firsts" of 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

    Spear & Sons Limited. Our scrabble word finder and scrabble cheat word builder is not associated with the Scrabble brand - we merely provide help for players of the official Scrabble game. All intellectual property rights to the game are owned by respective

  2. #2 peak

    it, “Cabby.” This is Cabby’s story:⏎To stay tuned on what Fant, and the rest of the Garden’s conservation scientists are doing, check out the latest news at chicagobot

  3. #3 peak

    iconic. Without it, Simon &- Garfunkel would not feel nearly as groovy. And just try⏎picturing Woody Allen’s Manhattan without that seeing him and Diane Keaton silhouetted against its criss-crossing

  4. #4 peak

    persons of another such description are not made subject or are accorded privileges or advantages<|endoftext|>Tom Standage, digital editor at the Economist, author of A History of the World in 6 Glasses and Writing on the Wall: Social Media - The First

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. L6:61 suppresses
  2. L8:636 suppresses
  3. L9:24 excites "the" before a named organisation or title
  4. L5:125 excites "of" after all three, both, ten
  5. L12:435 suppresses
  6. L4:52 excites
  7. L15:470 suppresses the word 'this' after a preposition
  8. L7:2164 suppresses period inside country abbreviations
  9. L3:344 excites
  10. L7:111 suppresses
  11. L9:82 suppresses
  12. L11:1168 excites
  13. L10:167 suppresses
  14. L4:968 excites
  15. L10:29 suppresses
  16. L14:169 suppresses
  17. L14:2291 suppresses
  18. L0:26 suppresses
  19. L0:774 suppresses
  20. L5:1747 suppresses
  21. L14:2846 suppresses
  22. L13:975 excites
  23. L5:1172 excites
  24. L3:291 excites

Feeds later layers

  1. L17:1154 weakly suppresses
  2. L23:750 weakly excites
  3. L25:542 weakly suppresses
  4. L27:2391 weakly suppresses
  5. L27:829 weakly excites
  6. L18:2670 weakly excites
  7. L18:1976 weakly suppresses
  8. L18:1354 weakly excites
  9. L18:1399 weakly suppresses
  10. L21:2665 weakly excites
  11. L19:2163 weakly suppresses
  12. L17:1638 weakly suppresses
  13. L19:518 weakly suppresses
  14. L17:1336 weakly excites
  15. L18:2926 weakly excites
  16. L21:713 weakly excites
  17. L26:927 weakly suppresses
  18. L22:2451 weakly excites
  19. L19:1628 weakly excites
  20. L26:706 weakly excites
  21. L18:517 weakly suppresses
  22. L22:1659 weakly excites
  23. L21:847 weakly excites
  24. L20:2161 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.

2 values between -8.82 and -8.471 values between -8.47 and -8.131 values between -7.78 and -7.433 values between -7.43 and -7.091 values between -6.74 and -6.394 values between -6.05 and -5.705 values between -5.70 and -5.354 values between -5.35 and -5.014 values between -5.01 and -4.6610 values between -4.66 and -4.3114 values between -4.31 and -3.9710 values between -3.97 and -3.6215 values between -3.62 and -3.2712 values between -3.27 and -2.9321 values between -2.93 and -2.5824 values between -2.58 and -2.2328 values between -2.23 and -1.8928 values between -1.89 and -1.5442 values between -1.54 and -1.1981 values between -1.19 and -0.85123 values between -0.85 and -0.50448 values between -0.50 and -0.152364 values between -0.15 and 0.19569 values between 0.19 and 0.5452 values between 0.54 and 0.8913 values between 0.89 and 1.2311 values between 1.23 and 1.585 values between 1.58 and 1.932 values between 1.93 and 2.274 values between 2.27 and 2.621 values between 2.62 and 2.972 values between 2.97 and 3.312 values between 3.31 and 3.661 values between 4.35 and 4.701 values between 4.70 and 5.05offnegativepositiveoften

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. 崮
  2. Users
  3. spons
  4. forEach
  5. herits
  6. 都市报
  7. poi
  8. 蚓
  9. AMPL
  10. sponsored

Suppresses

  1. ,{⏎
  2. например
  3. rí
  4. -basic
  5. 门
  6. <![
  7. <![
  8. <(
  9. //{⏎
  10. andbox