Neuron Explorer

Qwen3-0.6B

colon in a title and subtitle

A neuron in layer 1 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 colon that separates a book, article or report title from its subtitle, such as "The Value of Everything: Making and Taking...". Turning it up did not produce a clearly related change in the writing.

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

    life. But what’s being left out?⏎During one conversation at Food Secure Canada 8th National Assembly in Halifax, which was titled “Waves of Change: Sustainable Food for All,” community developers and food justice advocates, Aja Peterson and

  2. #2 peak

    magazine's "Lay of the Land."⏎And for background on the coal industry that's pushing this bad idea, download our report "The Dirty Truth About Coal: Why Yesterday's Technology Should Not Be Part of Tomorrow's Energy Future."⏎Make Mine

  3. #3 peak

    .S. soldiers.⏎Text adapted from “The Fight for Equal Rights: A Recruiting Poster for Black Soldiers in the Civil War” in the February

  4. #4 peak

    Revolution on Local Governments⏎- The Shale Revolution and Water Quality⏎Common Resources — Feb 7, 2018⏎The Facts on Fracking: A Podcast with Daniel Raimi⏎In this podcast, RFF’s Daniel

  5. #5 peak

    … May 04, 2010 · LEDC flooding case study Bangladesh 1998 1. LEDC Flooding Case Study: Bangladesh 1998 2. Between July-September 1

  6. #6 peak

    The Value of Everything: Making and Taking in the Global Economy by Mariana Mazzucato. Copyright

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: I need to be careful with my own ideas, and I need to be careful with my own thoughts, and I need to be careful with my own thoughts, and I need to be careful with my own thoughts, and I need to be

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

    Kids Listen Activity Podcast, Episode 9: Aaron's World Presents Origin of Dinosaurs⏎Aaron’s World is a fun and imaginative science-themed audio

  2. #2 peak

    11.1 Information Library|⏎- load exponent of a floating point number⏎c99 [ flag... ] file... -lm [ library... ] #include <math.h> double ldexp(double x, int exp);⏎float ldexp

  3. #3 peak

    bursts of colour as the different wild flowers start appearing with the summer rains. I recently saw a beautiful red coloured flower up on the high grasslands, Tritoniopsis caffra. Which has sword like leaves and vibrant red asymmetrical flowers that

  4. #4 peak

    shod feet touch the ground are apt to come home with painful stings inflicted by a tiny insect with a long name: Solenopsis invicta Buren.⏎Nicknamed RIFA, for red imported fire ant

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. L0:183 weakly suppresses
  2. L0:300 weakly excites
  3. L0:902 weakly suppresses
  4. L0:965 weakly excites
  5. L0:170 weakly excites
  6. L0:1530 weakly suppresses
  7. L0:2926 weakly excites
  8. L0:3026 weakly suppresses
  9. L0:1301 weakly suppresses
  10. L0:905 weakly suppresses
  11. L0:1748 weakly excites
  12. L0:120 weakly excites
  13. L0:849 weakly excites
  14. L0:2069 weakly excites
  15. L0:1963 weakly excites
  16. L0:989 weakly excites
  17. L0:1224 weakly suppresses
  18. L0:2113 weakly suppresses
  19. L0:2065 weakly suppresses
  20. L0:907 weakly suppresses
  21. L0:100 weakly excites first digit '1' of a year
  22. L0:1 weakly suppresses
  23. L0:866 weakly suppresses
  24. L0:1898 weakly excites

Feeds later layers

  1. L2:650 weakly suppresses
  2. L16:1518 weakly suppresses
  3. L25:1020 weakly excites
  4. L19:1557 weakly suppresses
  5. L18:2331 weakly suppresses
  6. L7:599 weakly suppresses
  7. L3:641 weakly excites
  8. L7:301 weakly excites
  9. L5:2292 weakly suppresses
  10. L24:1611 weakly suppresses
  11. L14:1786 weakly excites
  12. L23:2416 weakly suppresses
  13. L6:72 weakly suppresses
  14. L11:2606 weakly suppresses
  15. L11:2450 weakly suppresses
  16. L24:1481 weakly excites
  17. L23:2308 weakly excites
  18. L26:1991 weakly excites
  19. L23:1453 weakly suppresses
  20. L19:1212 weakly suppresses
  21. L8:1720 weakly excites
  22. L18:2121 weakly suppresses
  23. L22:605 weakly suppresses
  24. L25:1465 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 -0.60 and -0.551 values between -0.55 and -0.502 values between -0.50 and -0.451 values between -0.45 and -0.402 values between -0.40 and -0.353 values between -0.35 and -0.3012 values between -0.30 and -0.2522 values between -0.25 and -0.20107 values between -0.20 and -0.16190 values between -0.16 and -0.11393 values between -0.11 and -0.06897 values between -0.06 and -0.011195 values between -0.01 and 0.04635 values between 0.04 and 0.09272 values between 0.09 and 0.14105 values between 0.14 and 0.1932 values between 0.19 and 0.2422 values between 0.24 and 0.294 values between 0.29 and 0.342 values between 0.34 and 0.394 values between 0.39 and 0.441 values between 0.44 and 0.491 values between 0.49 and 0.531 values between 0.53 and 0.581 values between 1.22 and 1.271 values between 1.27 and 1.321 values between 1.32 and 1.37offnegativepositiveoften

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. EDIUM
  2. 家企业
  3. gie
  4. structions
  5. erview
  6. gies
  7. Husband
  8. <Entry
  9. hubby
  10. utations

Suppresses

  1. ing
  2. imir
  3. ibr
  4. as
  5. nodeName
  6. bald
  7. Notícias
  8. .getNode
  9. nodeName
  10. node