Neuron Explorer

Qwen3-0.6B

'it' in 'as it sounds' idioms

A neuron in layer 16 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron lights up on the word 'it' (or a similar word) in phrases like 'exactly what it sounds like', 'the way it is expected' and 'as it should be'. Turning it up made the model write unrelated text, so the phrase pattern is the main clue.

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

    We are very excited about the Bay Area Priority Conservation Area (PCA) program’s new urban greening designation.⏎Like PCAs, urban greening means exactly what it sounds like—improving our quality of life through trails, parks, street

  2. #2 peak

    system in order to detect bugs in the product so that they get fixed. Testing is an integral part of QC as it helps demonstrate that the product runs the way it is expected and designed for.⏎To summarize, think of everything as an assembly line

  3. #3 peak

    was a wave that brought in immigrants from Asia and Africa.⏎And so, there were times in our history where the U.S. was just not the country it was supposed to be.⏎In that sense, the story of how our nation has come

  4. #4 peak

    liver disease is a condition that can have a significant impact on quality of life, yet it is not something we hear about very often. The condition is exactly as it sounds. Fatty liver disease develops when an excessive amount of fat builds up in

  5. #5 peak

    00. Within four months, these men<|endoftext|>Our discourse by now may have proved to the reader that the⏎British Empire is not as mythical as the propaganda would have us believe.⏎We quote from THE NATIONAL MESSAGE, of London, England,

  6. #6 peak

    actual glimpse of its mystic history. From ancient buildings to the modern wonders Delhi offers anything one can desire for. The rulers who have ruled India from Delhi have left their influence in the

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: if the following, if it is, and if it, and if it, and if it, and if it, and if it, and if it, and if it, and if it, and if and only if, and if

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

    .⏎Discovering this now, we object to this language – we are same-sex attracted, and we wonder at how this woolly description was approved, and on whose recommendation. However, in mitigation, this version

  2. #2 peak

    grief. Now in whose grief does a newborn child weep? That’s the question, which can be

  3. #3 peak

    no evidence indicating that if you eat after a certain time, you are guaranteed to gain weight. Based on your body type and height, your body needs a specific number of calories in order to maintain your present weight. If you consume all of those

  4. #4 peak

    of the fasting month Ramadan.⏎- Laylat al-Qaddr (Night of Measure): ≈ 27th Ramadan*⏎Muslims believe that it was on this night that God has first send down the first verses of the Quran via the angel

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. L13:1847 suppresses
  2. L14:925 suppresses
  3. L15:1132 excites
  4. L13:1770 suppresses
  5. L15:1407 excites
  6. L15:1923 excites
  7. L13:2980 excites
  8. L15:835 suppresses
  9. L12:1255 suppresses
  10. L13:2715 suppresses
  11. L8:2827 excites
  12. L14:1854 suppresses
  13. L13:1147 excites
  14. L15:836 excites
  15. L14:2886 suppresses
  16. L13:2267 excites
  17. L13:2376 excites
  18. L7:934 excites
  19. L12:1218 excites
  20. L12:2578 suppresses
  21. L12:870 suppresses
  22. L14:1779 excites
  23. L14:2359 excites
  24. L6:894 excites

Feeds later layers

  1. L18:780 suppresses
  2. L18:1983 suppresses
  3. L17:1528 suppresses
  4. L21:1420 excites
  5. L17:1689 suppresses
  6. L18:2459 suppresses
  7. L17:2942 suppresses
  8. L17:2044 suppresses
  9. L18:2328 suppresses
  10. L18:1829 weakly suppresses
  11. L20:1388 weakly suppresses
  12. L18:2027 weakly suppresses
  13. L18:2157 weakly suppresses
  14. L18:1577 weakly suppresses
  15. L18:1410 weakly suppresses
  16. L18:483 weakly excites
  17. L19:2637 weakly suppresses
  18. L18:517 weakly suppresses
  19. L17:1477 weakly excites
  20. L20:2137 weakly suppresses
  21. L18:2379 weakly excites
  22. L17:2696 weakly suppresses
  23. L17:2057 weakly suppresses
  24. L17:2504 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.

1 values between -5.25 and -5.061 values between -4.48 and -4.292 values between -3.53 and -3.332 values between -3.14 and -2.951 values between -2.76 and -2.573 values between -2.57 and -2.384 values between -2.38 and -2.185 values between -2.18 and -1.994 values between -1.99 and -1.805 values between -1.80 and -1.619 values between -1.61 and -1.4220 values between -1.42 and -1.2228 values between -1.22 and -1.0349 values between -1.03 and -0.8457 values between -0.84 and -0.6577 values between -0.65 and -0.46180 values between -0.46 and -0.27593 values between -0.27 and -0.071919 values between -0.07 and 0.12538 values between 0.12 and 0.31191 values between 0.31 and 0.5074 values between 0.50 and 0.6950 values between 0.69 and 0.8830 values between 0.88 and 1.0818 values between 1.08 and 1.2723 values between 1.27 and 1.469 values between 1.46 and 1.655 values between 1.65 and 1.847 values between 1.84 and 2.042 values between 2.04 and 2.231 values between 2.23 and 2.42offnegativepositiveoften

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. ulk
  2. ul
  3. 实务
  4. ologies
  5. chồng
  6. urers
  7. oint
  8. trak
  9. ост
  10. index

Suppresses

  1. 的产品
  2. 的商品
  3. したもの
  4. 的东西
  5. 的女儿
  6. 的房子
  7. 的文章
  8. 指导下
  9. 的时间里
  10. 的结果