Neuron Explorer

Qwen3-0.6B

prepositions in how-to location phrases

A neuron in layer 23 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on small connecting words such as 'in', 'to', 'the' or 'your' in step-by-step instructions, especially where the text says where something is or where to go. Turned up, the model produced long runs of the letter C.

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

    files, the syntax in configuration files is case-sensitive.⏎The machine configuration file, Machine.config, contains settings that apply to an entire computer. This file is located in the %runtime

  2. #2 peak

    306 Survey Data.xlsx file from BlackBoard. Immediately save this file to your H: drive or memory stick. For purposes of this assignment consider these to be sample data

  3. #3 peak

    Files?⏎We can zip and unzip files in Windows 10 easily. To zip files:⏎Step 1: Go to File Explorer and go to the files and media you want to zip.⏎Step

  4. #4 peak

    new instance.⏎Windows MSI package installers create and set up the data directory that the installed server uses, but also create a pristine “template” data directory named⏎dataunder the installation directory. After an installation has been performed using an MSI package

  5. #5 peak

    Step⏎Finally, you have followed all the steps, you will see one shortcut on the desktop. Cut it and paste it at the following path⏎Now open Run (ctrl+R) and type your command name(Here

  6. #6 peak

    , we first have to mount it. Click on Places and then the location that you want to store the image on – in our case, a 136GB internal drive.⏎Open a terminal window (Applications >

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 question is, how much do we know about the number of human human human human human human human human human human human human human human human human human human human human human human human human human human human human human human human human human human human

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

    1/4 is 1/2 of 1/2<|endoftext|>Messier 52 (also known as M 52 or NGC 7654) is an open cluster in the Cassiopeia constellation.

  2. #2 peak

    the cluster, with the brightest member being magnitude 11.⏎Messier 52 is evaluated at about 35 million years old.⏎NGC 7635, also called the Bubble Nebula, Sharpless 16

  3. #3 peak

    eldovich (1967) and Field (1974) and has been clearly incorporated in the well-known book by Weinberg (1972). Some more recent books also address this analysis (Kolb and Turner

  4. #4 peak

    CE. He is also known for being the conqueror of Jerusalem.⏎What did Titus do?⏎Titus commanded a Roman legion in Judaea. In 70 CE he led a campaign that culminated in the capture and destruction of Jerusalem

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. L22:807 excites
  2. L22:2104 excites
  3. L19:2357 suppresses
  4. L19:850 excites
  5. L22:1560 suppresses
  6. L4:1679 weakly excites
  7. L22:983 weakly excites
  8. L22:1065 weakly suppresses
  9. L2:1407 weakly suppresses
  10. L1:2569 weakly suppresses
  11. L20:2394 weakly suppresses
  12. L20:2585 weakly suppresses
  13. L9:394 weakly excites
  14. L0:1498 weakly excites
  15. L22:1928 weakly excites
  16. L15:889 weakly suppresses end of text after an abrupt cut
  17. L13:661 weakly suppresses
  18. L11:2684 weakly suppresses
  19. L18:1317 weakly suppresses
  20. L22:1649 weakly excites
  21. L22:1247 weakly excites
  22. L16:1446 weakly excites
  23. L7:2413 weakly excites
  24. L11:2153 weakly suppresses

Feeds later layers

  1. L27:629 strongly excites
  2. L27:495 strongly excites
  3. L24:658 strongly excites
  4. L24:485 excites
  5. L26:519 excites
  6. L26:297 excites
  7. L25:1154 excites
  8. L26:1427 excites
  9. L25:616 weakly excites
  10. L26:1055 weakly excites
  11. L27:552 weakly excites
  12. L26:539 weakly suppresses
  13. L26:278 weakly excites
  14. L25:400 weakly excites
  15. L27:853 weakly suppresses
  16. L27:817 weakly suppresses
  17. L24:1819 weakly suppresses
  18. L26:2168 weakly excites
  19. L24:2138 weakly suppresses comma before 'right?' tag
  20. L26:2339 weakly suppresses
  21. L24:1376 weakly excites
  22. L25:1344 weakly suppresses
  23. L24:1383 weakly suppresses
  24. L25:2288 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 -14.97 and -14.511 values between -12.67 and -12.211 values between -11.75 and -11.291 values between -6.69 and -6.231 values between -6.23 and -5.772 values between -5.77 and -5.311 values between -4.39 and -3.931 values between -3.93 and -3.472 values between -3.47 and -3.011 values between -3.01 and -2.554 values between -2.55 and -2.092 values between -2.09 and -1.634 values between -1.63 and -1.1737 values between -1.17 and -0.72346 values between -0.72 and -0.262912 values between -0.26 and 0.21502 values between 0.21 and 0.6771 values between 0.67 and 1.1312 values between 1.13 and 1.593 values between 1.59 and 2.042 values between 2.04 and 2.501 values between 2.96 and 3.42offnegativepositiveoften

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. C
  2. C
  3. _C
  4. ,C
  5. *C
  6. :C
  7. .C
  8. +C
  9. >C
  10. =C

Suppresses

  1. óż
  2. ialect
  3. 白银
  4. trench
  5. LOT
  6. .scss
  7. иона
  8. ancybox
  9. ュ
  10. ТЬ