Neuron Explorer

Qwen3-0.6B

'prone to' and similar phrases

A neuron in layer 0 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 often lights up on the words 'to' after 'prone' (as in 'prone to errors'), and on a few other unrelated words. When it was turned up, the written text did not clearly change in a way that relates to this.

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

    California is at about 48%.⏎The Garbage Crisis⏎The world has changed a lot in the past century. From individually packaged food servings to disposable diapers, more garbage is generated now than ever before. The average American discards seven

  2. #2 peak

    nutrient depletion and to increase organic matter, which makes it less prone to drought, flooding and water-logging. The project has helped reduce pressure on the

  3. #3 peak

    8, in the Nihon Shoki 日本書紀 (720) 9 and in the Hizen-no-kuni fudoki 肥前國風土記 (first part of the 8th century)

  4. #4 peak

    or the expression.⏎from WordNet 3.0 Copyright 2006 by Princeton University. All rights reserved.⏎- adj. somewhat ill or prone to illness⏎- adj. having or rising to a peak⏎According to the Twitter

  5. #5 peak

    affected by electrical noise emitted by gadgets like smartphones, TVs, computers, even vacuum cleaners and hairdryers, AM signals are prone to distortion and crackling.⏎That crackle apparently just doesn’t fly with luxury auto

  6. #6 peak

    ,570 and $65,260 a year.⏎Salary by Employment Sector⏎As of 2011, the highest-paid electricians were those who taught at business schools. These workers earned an average of $7

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 am going to take the first step. I am going to take the first step. I am going to take the first step. I am going to take the first step. I am going to take the first step. I am going

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

    only declared as:⏎void * memchr ( const void *, int, size_t );⏎instead of the two overloaded versions provided in C++.⏎/* memchr example */⏎int main ()⏎char * pch;⏎char str = "Example string";⏎pch

  2. #2 peak

    full list of Biblical Festivals.⏎Question: How many days should Passover be observed? Please include any references to your response that is indicated in the Torah.⏎Answer: Before I answer your question, as a word of background, you need to

  3. #3 peak

    . Numerous ancient Greeks believed that Iris served as a conduit between Earth and the divine, metaphorically represented

  4. #4 peak

    , the two advocates changed the course of American history.<|endoftext|>Question: What is it significance of the date passover falls on and the phase of the moon?⏎Answer: The date on which to observe Passover comes from the Torah -- it is

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

None: this is the first layer.

Feeds later layers

  1. L22:2664 weakly suppresses
  2. L12:2382 weakly suppresses
  3. L2:2627 weakly excites
  4. L18:1357 weakly excites
  5. L3:1030 weakly suppresses
  6. L14:685 weakly suppresses
  7. L10:352 weakly excites
  8. L16:2165 weakly excites
  9. L11:590 weakly suppresses
  10. L13:946 weakly suppresses
  11. L18:201 weakly excites
  12. L4:2800 weakly excites
  13. L13:2570 weakly suppresses
  14. L19:2093 weakly suppresses
  15. L1:892 weakly excites
  16. L6:1901 weakly suppresses
  17. L14:1157 weakly suppresses
  18. L10:1520 weakly excites
  19. L16:779 weakly excites
  20. L19:762 weakly suppresses
  21. L7:1886 weakly suppresses
  22. L24:562 weakly excites
  23. L9:200 weakly excites
  24. L6:2952 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.

3 values between -0.42 and -0.392 values between -0.34 and -0.311 values between -0.31 and -0.282 values between -0.28 and -0.2610 values between -0.26 and -0.2311 values between -0.23 and -0.2022 values between -0.20 and -0.1828 values between -0.18 and -0.1571 values between -0.15 and -0.12109 values between -0.12 and -0.10196 values between -0.10 and -0.07407 values between -0.07 and -0.04656 values between -0.04 and -0.011203 values between -0.01 and 0.01534 values between 0.01 and 0.04304 values between 0.04 and 0.07166 values between 0.07 and 0.0989 values between 0.09 and 0.1251 values between 0.12 and 0.1521 values between 0.15 and 0.186 values between 0.18 and 0.208 values between 0.20 and 0.234 values between 0.23 and 0.261 values between 0.28 and 0.311 values between 0.34 and 0.361 values between 0.45 and 0.471 values between 0.63 and 0.66offnegativepositiveoften

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. burg
  2. aleza
  3. iteur
  4. rome
  5. ertools
  6. uating
  7. idel
  8. minster
  9. cape
  10. ittel

Suppresses

  1. annot
  2. alcon
  3. queried
  4. searched
  5. Pag
  6. annotated
  7. ouples
  8. ancode
  9. sad
  10. chast