Neuron Explorer

Qwen3-0.6B

colon in clock times

A neuron in layer 25 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 colon in times of day, like 01:32 AM or 15:47, in forum posts and timestamps. Turning it down or up produced text that did not clearly relate to times, so the steering does not confirm the pattern.

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

    actorial.shtml I don't happen to think that the code there is easily understandable though...⏎January 10th, 2004 01:32 AM⏎Tim_axe, sorry to be pedantic, but if

  2. #2 peak

    C=0 and MAS[1:0]=10).⏎Article last edited on: 2008-09-09 15:47:35⏎Did you find this article helpful? Yes No⏎

  3. #3 peak

    of the Marquis de Armijo.<|endoftext|>Thread: FNG methods of attack question⏎05-06-2006, 05:45 PM #1⏎FNG methods of attack question⏎I read this

  4. #4 peak

    dream up the basic elements of the classroom of the future using design thinking. Great example of a school-business partnership.⏎- Design & Thinking Official Trailer (02:28)⏎Design & Thinking is a documentary released this year about design thinking.

  5. #5 peak

    2:51:40 EDT⏎- Martian rock arrangement not alien handiworkWed, 7 Jan 2009, 13:

  6. #6 peak

    Awareness of climate change] is<|endoftext|>- Created by: Luke Simons⏎- Created on: 06-04-12 19:17⏎1. Newton's model of the universe made four assumptions which were

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: what is the place of government in an environment in which government is supposed to do exactly what it is supposed to do? Well, in fact, government is supposed to do exactly what it is supposed to do. And yet, as government in

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

    Jurisdiction permits the homeowner to do everything that a licensed electrician can do, but the work must be performed under permit and the work must be inspected by a building department inspector. The documents that you will need to

  2. #2 peak

    increases.⏎The mechanical behavior of fractured rocks has directly impact on the stability of slope, tunnel and dam. etc. The research on creep shear strength of fractured rocks can provide

  3. #3 peak

    and to display results in near real-time. For at least 15 minutes on one or more days of the count, simply tally the numbers and kinds of birds you see. You can count from any location

  4. #4 peak

    options contract has the following properties or attributes:⏎- Expiration Date⏎- Symbol or Underlying⏎- Strike Price⏎- Type: Call or Put⏎The contract has an

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. L21:326 excites
  2. L23:660 suppresses
  3. L20:2736 suppresses
  4. L22:2408 suppresses
  5. L20:2316 excites
  6. L22:610 suppresses
  7. L18:1614 suppresses
  8. L24:800 weakly excites colon in clock times
  9. L23:1369 weakly excites the '0' in years like 2008
  10. L23:2791 weakly excites
  11. L23:231 weakly excites
  12. L14:631 weakly suppresses
  13. L19:920 weakly excites
  14. L19:2230 weakly excites
  15. L22:2254 weakly suppresses
  16. L9:804 weakly suppresses
  17. L11:1996 weakly excites
  18. L15:1815 weakly suppresses
  19. L23:2992 weakly suppresses unclear: first pieces of split words
  20. L23:585 weakly excites
  21. L7:1266 weakly excites
  22. L9:1719 weakly suppresses
  23. L0:672 weakly excites
  24. L12:408 weakly excites

Feeds later layers

  1. L27:496 suppresses
  2. L26:905 suppresses
  3. L27:763 suppresses
  4. L27:205 excites
  5. L27:572 suppresses
  6. L27:1087 suppresses
  7. L27:435 suppresses
  8. L27:1100 suppresses
  9. L27:132 excites
  10. L27:893 suppresses
  11. L26:233 weakly suppresses
  12. L27:628 weakly suppresses
  13. L27:1329 weakly suppresses
  14. L27:1232 weakly suppresses
  15. L26:294 weakly suppresses
  16. L27:391 weakly suppresses
  17. L26:976 weakly suppresses the '1' starting a century
  18. L26:580 weakly suppresses
  19. L27:773 weakly excites
  20. L27:1626 weakly suppresses
  21. L27:530 weakly suppresses
  22. L26:898 weakly suppresses
  23. L26:2380 weakly excites
  24. L26:1551 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 -19.42 and -18.421 values between -16.42 and -15.422 values between -15.42 and -14.421 values between -14.42 and -13.425 values between -13.42 and -12.426 values between -12.42 and -11.4214 values between -11.42 and -10.4215 values between -10.42 and -9.4217 values between -9.42 and -8.4236 values between -8.42 and -7.4230 values between -7.42 and -6.4286 values between -6.42 and -5.4296 values between -5.42 and -4.42162 values between -4.42 and -3.42211 values between -3.42 and -2.42341 values between -2.42 and -1.42580 values between -1.42 and -0.41990 values between -0.41 and 0.59554 values between 0.59 and 1.59290 values between 1.59 and 2.59190 values between 2.59 and 3.59101 values between 3.59 and 4.5962 values between 4.59 and 5.5933 values between 5.59 and 6.5936 values between 6.59 and 7.5915 values between 7.59 and 8.5913 values between 8.59 and 9.596 values between 9.59 and 10.591 values between 10.59 and 11.592 values between 11.59 and 12.594 values between 12.59 and 13.591 values between 14.59 and 15.591 values between 15.59 and 16.592 values between 16.59 and 17.592 values between 19.59 and 20.59offnegativepositiveoften

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. -bar
  2. Bars
  3. вая
  4. .hh
  5. ubble
  6. backup
  7. ilda
  8. entine
  9. 憩
  10. èles

Suppresses

  1. 8
  2. 6
  3. 7
  4. 8
  5. 6
  6. 七个
  7. 7
  8. eighty
  9. six
  10. eight