Neuron Explorer

Qwen3-0.6B

'th' in '20th century'

A neuron in layer 6 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 very strongly on the 'th' in '20th century' and similar phrases about the 1900s. Turning it up did not produce any obvious theme in the sample text, so the steering results don't add much. Its weak negative side is on 'Development' in 'U.S. Agency for International Development'.

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

    see, in some areas, city wells or fountains decorated with evergreens and with eggs also. This is a newer tradition, developed in the 20th century, though it uses old symbols of life – the idea of decorating or “

  2. #2 peak

    60s began an era of change that is still being worked out.⏎Readers can also take a more thematic view of American history in the 20th century. Here we will list only the major events and developments.⏎American<|endoftext|>What

  3. #3 peak

    of Knowledge. Open University Press.⏎José M. Sanchez-Ron (1985). Physics and Philosophy: Action at a Distance in 20th Century Physics. Theoria 1 (2):439-45

  4. #4 peak

    American power that today even the smallest European state can enjoy its national identity and independence in unprecsented security. Domestically America in the 20th century gradually implemented a series of progressiveand lineral reforms designed to widen the opportunity

  5. #5 peak

    finding a particular solution: this leads into some of the deepest areas in all of mathematics, both conceptually and in terms of technique.⏎In the 20th century, algebraic geometry has split into several subareas.⏎- The main stream

  6. #6 peak

    aint-Laurent, known as Yves Saint Laurent was a French fashion designer, one of the greatest names in French fashion in the 20th century. In 1985, Caroline Rennolds Milbank wrote

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 want to know that I am in the wrong. I don't know that I am in the wrong. I don't know that I am in the wrong. I'm in the wrong." So, the problem is that I have

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

    , according to Wilson Center Senior Scholar Murray Feshbach and Research Associate Cristina Galvin in a recently released report underwritten by the U.S. Agency for International Development.⏎"The situation we face in China, India, and Russia bears alarming similarities to

  2. #2 peak

    , and promoting sustainable economic development.⏎U.S. Assistance to Solomon Islands⏎In Solomon Islands and across the Pacific Islands region, the U.S. Agency for International Development (USAID) supports programs that help communities adapt to the negative impacts of global

  3. #3 peak

    hidden faults. These hidden faults are pathways for potential CO2 leaks, at least near the surface, over the entire region. The methods being developed use airborne hyperspectral and multi-spectral high-resolution imagery and very high resolution (0.<|endoftext|>Components

  4. #4 peak

    staff element that is part of the 45th Space Wing staff at Patrick Air Force Base, Fl.⏎The commander of U.S. Strategic Command (USSTRATCOM) was the DoD Manager for Manned Space Flight Support Operations.

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. L4:1594 weakly suppresses
  2. L1:616 weakly suppresses
  3. L4:2355 weakly suppresses
  4. L3:2162 weakly suppresses
  5. L4:2848 weakly suppresses
  6. L4:2446 weakly suppresses
  7. L1:2570 weakly suppresses
  8. L5:2801 weakly excites
  9. L1:1901 weakly excites
  10. L4:1744 weakly excites
  11. L5:28 weakly excites
  12. L4:1889 weakly excites
  13. L2:2454 weakly excites
  14. L1:1870 weakly suppresses
  15. L4:2792 weakly suppresses
  16. L5:158 weakly suppresses
  17. L5:979 weakly suppresses
  18. L3:2074 weakly excites
  19. L1:1500 weakly excites
  20. L0:2457 weakly excites
  21. L1:1757 weakly suppresses
  22. L0:658 weakly suppresses
  23. L0:1768 weakly suppresses
  24. L3:162 weakly suppresses

Feeds later layers

  1. L9:1212 excites
  2. L9:1480 excites
  3. L22:906 excites
  4. L7:1012 excites
  5. L19:2030 excites
  6. L8:795 weakly excites
  7. L11:2104 weakly excites
  8. L8:360 weakly suppresses
  9. L8:1331 weakly excites
  10. L8:261 weakly suppresses
  11. L10:952 weakly suppresses
  12. L11:997 weakly suppresses
  13. L9:3024 weakly excites
  14. L24:904 weakly excites
  15. L11:1128 weakly suppresses
  16. L18:2699 weakly excites
  17. L11:1211 weakly excites
  18. L7:279 weakly excites
  19. L21:2809 weakly excites
  20. L10:1142 weakly suppresses
  21. L7:3059 weakly excites
  22. L25:2034 weakly excites
  23. L21:2983 weakly excites
  24. L20:2094 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 -1.30 and -1.181 values between -1.18 and -1.054 values between -0.80 and -0.684 values between -0.68 and -0.5514 values between -0.55 and -0.4372 values between -0.43 and -0.30349 values between -0.30 and -0.18966 values between -0.18 and -0.051580 values between -0.05 and 0.07666 values between 0.07 and 0.20169 values between 0.20 and 0.3238 values between 0.32 and 0.4516 values between 0.45 and 0.575 values between 0.57 and 0.702 values between 0.70 and 0.823 values between 0.82 and 0.951 values between 0.95 and 1.074 values between 1.07 and 1.204 values between 1.20 and 1.321 values between 1.32 and 1.451 values between 1.82 and 1.951 values between 2.07 and 2.201 values between 2.57 and 2.702 values between 2.70 and 2.821 values between 2.95 and 3.071 values between 3.45 and 3.571 values between 3.57 and 3.70offnegativepositiveoften

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. Ạ
  2. lửa
  3. 舰队
  4. generation
  5. 的城市
  6. 的知识
  7. odelist
  8. идент
  9. endid
  10. tone

Suppresses

  1. 笼
  2. iliar
  3. @js
  4. ugging
  5. ucus
  6. .owl
  7. bling
  8. ismatic
  9. cir
  10. iliate