Neuron Explorer

Qwen3-0.6B

"in" before "good" condition

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 fires on the word "in" when it is about to be followed by "good", as in "keep our bodies in good working order" or "in good repair". Turning it up made the writing repeat the word "good" again and again, which fits what it is tuned to predict.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    was disassembled. A great deal of the vessel’s wood had rotted significantly by that time, but much of the planking was in good condition despite having been under water all those years. Leaching from the ship

  2. #2 peak

    the most amazing diet (which is, of course, extremely important), but if our nervous system is out of balance, even the best diet won’t keep us in a healthy state.⏎Our autonomic nervous system plays a role in maintaining many functions

  3. #3 peak

    . Retrieved from https://digitalcommons.pvamu.edu/pvamu-theses/395<|endoftext|>The Top Reasons to Take Vitamins⏎Keep Our Bodies in Good Working Order⏎Vitamins work hard to keep our bodies functioning properly and

  4. #4 peak

    directly to the repositories.<|endoftext|>Energy Saving Tips⏎Your Home’s Energy Use⏎Tips to Reduce Energy Consumption and Lower Energy Bills⏎Keep your heating-cooling equipment in good repair. Equipment operating at peak efficiency can reduce fuel consumption.⏎- Insulate

  5. #5 peak

    laboratory, known as Wardenclyffe and envisioned by Tesla as a prototype for a stupendous world communications and wireless electric system, is still standing and in relatively good shape only yards from Route 25A, a major state highway.⏎

  6. #6 peak

    tuberculosis.⏎Health care cost reductions⏎“The best medicine is having to use no medicine” means that if you live an active and healthy life, then you will be in

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 does it mean to make a system in the community? In other words, how does community come in to play in making extended home/community communities? My first person point is that the community doesnxicxicxicxicxicxicxicxicxic

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

    the significance of the life and music captured in the recordings. The article contains a wealth of information on Morton’s early life, his uneven career, his incisive and wide-ranging views on jazz and related musical forms, and his central

  2. #2 peak

    00 members, ostensibly for the state party organization's attacks on religion and repudiation of "immediate demands." Michigan Socialist Oakley C. Johnson was incensed, charging in the pages of The Proletarian that the NEC had "

  3. #3 peak

    was unmistakable – and not just in Atlanta, the most aviation-oriented city in the former Confederacy. Jim Crow Terminals, Anke Ortlepp’s incisive new history of the desegregation of American airports, demonstrates that almost

  4. #4 peak

    In the next few years we will see -mandated by legislation in 2007 -the elimination of the incandenscent light bulb as we know it. By the year 20

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. L24:1668 suppresses
  2. L22:1093 suppresses
  3. L16:1860 suppresses
  4. L19:2406 excites
  5. L20:1574 excites
  6. L21:1160 weakly excites 'is' after a subject noun about likelihood
  7. L19:2685 weakly suppresses
  8. L23:2991 weakly excites
  9. L22:2526 weakly suppresses
  10. L20:1919 weakly excites
  11. L24:1843 weakly excites
  12. L0:98 weakly suppresses
  13. L22:151 weakly suppresses
  14. L22:554 weakly suppresses
  15. L17:812 weakly excites
  16. L20:669 weakly suppresses
  17. L23:143 weakly suppresses
  18. L18:2240 weakly suppresses
  19. L21:7 weakly excites mixed, weak code and reference hits
  20. L6:2346 weakly excites
  21. L2:588 weakly suppresses
  22. L24:80 weakly excites
  23. L23:2881 weakly excites
  24. L6:191 weakly excites

Feeds later layers

  1. L27:1897 excites
  2. L27:1650 excites
  3. L26:2816 weakly suppresses
  4. L26:2527 weakly excites
  5. L26:2582 weakly excites
  6. L27:1830 weakly suppresses
  7. L26:2308 weakly excites
  8. L26:1421 weakly suppresses
  9. L27:2629 weakly excites
  10. L26:792 weakly excites
  11. L26:1191 weakly suppresses
  12. L27:1398 weakly suppresses
  13. L27:1298 weakly suppresses
  14. L26:567 weakly excites
  15. L26:2317 weakly suppresses
  16. L26:3021 weakly suppresses
  17. L26:1492 weakly excites
  18. L27:1076 weakly excites
  19. L26:545 weakly excites
  20. L27:2205 weakly suppresses
  21. L27:2939 weakly excites
  22. L26:789 weakly excites
  23. L27:2910 weakly excites
  24. L26:453 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 -13.29 and -12.251 values between -12.25 and -11.204 values between -11.20 and -10.162 values between -10.16 and -9.125 values between -9.12 and -8.083 values between -8.08 and -7.0414 values between -7.04 and -5.9913 values between -5.99 and -4.9526 values between -4.95 and -3.9134 values between -3.91 and -2.8785 values between -2.87 and -1.82200 values between -1.82 and -0.782406 values between -0.78 and 0.26901 values between 0.26 and 1.3094 values between 1.30 and 2.3537 values between 2.35 and 3.3920 values between 3.39 and 4.4318 values between 4.43 and 5.4713 values between 5.47 and 6.5210 values between 6.52 and 7.563 values between 7.56 and 8.604 values between 8.60 and 9.644 values between 9.64 and 10.691 values between 10.69 and 11.733 values between 12.77 and 13.811 values between 14.86 and 15.901 values between 15.90 and 16.941 values between 24.24 and 25.282 values between 27.36 and 28.41offnegativepositiveoften

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. good
  2. 好
  3. good
  4. Good
  5. Good
  6. bad
  7. GOOD
  8. 好的
  9. -good
  10. _good

Suppresses

  1. чки
  2. arence
  3. ifecycle
  4. isable
  5. icts
  6. hurst
  7. иков
  8. ises
  9. appen
  10. ới