Neuron Explorer

Qwen3-0.6B

"the" right after an action verb

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 fires on "the" when it comes right after an action word in instructions, such as "Divide the total", "Use the method" or "set the variable". The effect on the model's writing was not clear when turned up.

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

    the total weight of the vehicle and its maximum load.⏎- Add an additional 25 percent of the total weight to allow a margin of error.⏎- Divide the total weight including the additional 25 percent by the number of wheels supporting the

  2. #2 peak

    string, it is returned. Otherwise a case-insensitive search is performed, and if a DataTable name is found that matches this search, it is returned.⏎Use the Contains method to determine whether a

  3. #3 peak

    - Your second subroutine should be called getVolumeOfCuboid(), take three parameters called width, length and height and apply the relevant formula to calculate and return the volume of a cuboid based on the given dimensions.<|endoftext|>Approximately 3

  4. #4 peak

    over it is more convenient to create a script that can run all the commands with one message from you.⏎% filename simple.m⏎ini=1.; %set the variable ini to 1.⏎y0=ini; %set the initial condition

  5. #5 peak

    Add “c:\Borland\Bcc55\bin” to the existing path⏎- Create a bcc32.cfg file which will set the compiler options for the Include and Lib

  6. #6 peak

    concatenated.⏎You can create a RAID 5 volume without having to “zero out” the data blocks. To do so, do one of the following:⏎Use the metainit command with the -k option. The -k option recreates

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 need for a good education system. The need for a good education system. And so many people, including my parents, including my friends, including my family, are so many people, including my friends, including my family, are so

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

    investors will also skip companies who don’t engage in socially responsible behavior (like coffee/restaurant companies who don’t use fair trade beans, for

  2. #2 peak

    to bring man's esteemed friend into the postindustrial age of the genome. Compared with the Human Genome Project, the

  3. #3 peak

    a generation of socially responsible society.⏎Third, the schools/universities, by having a standard curriculum,

  4. #4 peak

    glimpse into the academic world and its view of religion in general and Hinduism in particular. The

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. L21:1608 weakly suppresses
  2. L25:2547 weakly excites
  3. L16:622 weakly suppresses
  4. L8:2125 weakly suppresses
  5. L7:593 weakly excites
  6. L9:1542 weakly suppresses
  7. L12:122 weakly excites unclear: word before a clause continues
  8. L15:1843 weakly suppresses
  9. L25:2074 weakly suppresses
  10. L13:1996 weakly suppresses
  11. L4:941 weakly excites
  12. L10:2703 weakly suppresses
  13. L4:499 weakly excites
  14. L20:2634 weakly excites
  15. L23:2676 weakly excites
  16. L7:725 weakly excites
  17. L15:1689 weakly suppresses
  18. L25:1132 weakly excites
  19. L13:483 weakly suppresses
  20. L4:2259 weakly suppresses
  21. L23:541 weakly suppresses
  22. L6:1572 weakly excites
  23. L24:1343 weakly excites
  24. L9:903 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 -1.16 and -1.121 values between -1.12 and -1.081 values between -1.04 and -1.002 values between -1.00 and -0.962 values between -0.96 and -0.921 values between -0.92 and -0.884 values between -0.88 and -0.845 values between -0.84 and -0.806 values between -0.80 and -0.767 values between -0.76 and -0.7212 values between -0.72 and -0.6810 values between -0.68 and -0.649 values between -0.64 and -0.6010 values between -0.60 and -0.568 values between -0.56 and -0.528 values between -0.52 and -0.4812 values between -0.48 and -0.448 values between -0.44 and -0.406 values between -0.40 and -0.3614 values between -0.36 and -0.3212 values between -0.32 and -0.2818 values between -0.28 and -0.2422 values between -0.24 and -0.2033 values between -0.20 and -0.1647 values between -0.16 and -0.1286 values between -0.12 and -0.08238 values between -0.08 and -0.041084 values between -0.04 and -0.001510 values between -0.00 and 0.04490 values between 0.04 and 0.08136 values between 0.08 and 0.1261 values between 0.12 and 0.1624 values between 0.16 and 0.2010 values between 0.20 and 0.242 values between 0.24 and 0.282 values between 0.28 and 0.323 values between 0.32 and 0.362 values between 0.36 and 0.401 values between 0.40 and 0.44offnegativepositiveoften

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. PackageName
  2. setC
  3. pheric
  4. ServiceImpl
  5. -Pack
  6. importing
  7. packageName
  8. sims
  9. realistically
  10. pics

Suppresses

  1. __(*
  2. itet
  3. or
  4. ift
  5. autom
  6. minster
  7. nor
  8. /qt
  9. 座
  10. entral