Neuron Explorer

Qwen3-0.6B

"and" joining abstract nouns

A neuron in layer 4 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 lights up on the word "and" when it links two formal, abstract nouns, like "development and dissemination" or "empowerment and liberation." When it was turned up, the model's writing became repetitive and didn't stay on topic, so the steering output doesn't show a clear theme.

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

    for the most vulnerable in society’⏎Centacare Catholic Family Services, Australia⏎Through the promotion of societal change, problem-solving in human relationships and the empowerment and liberation of people to enhance their well-being, social services provides a framework to work

  2. #2 peak

    Peace and Development highlights the important role of science in society and the need to engage the wider public in debates on emerging scientific issues. It also underlines the importance and relevance of science in our daily lives.⏎By linking science more closely with society,

  3. #3 peak

    thetic organisms to support life.⏎Fundamental Concept 5h.⏎Tides, waves and predation cause vertical zonation patterns along the shore, influencing the distribution and diversity

  4. #4 peak

    their daily life.⏎What is Computer Science?⏎Computer Science (CS) is the study and application of software, hardware, and algorithms in solving today's real-life problems.

  5. #5 peak

    papyrifera_), of _hau_ (_Hilasens tiliasens_), etc.<|endoftext|>The American Meteorological Society⏎) promotes the development and dissemination of information and education on the atmospheric⏎Atmospheric sciences is an umbrella

  6. #6 peak

    Values⏎In performing its mission, ASME adheres to these core values:⏎•Embrace integrity and ethical conduct⏎•Embrace diversity and respect the dignity and culture of all people⏎•Nurture and treasure the environment and our natural

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: - 2023 # Math # Coding # 2023 # Math # Coding # 2023 # Math # 2023 # Math #

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

    to be met, and health service providers will need to plan for a significant increase in obesity-linked comorbidities.⏎The researchers were particularly interested to know what to expect in the way of co-morbidities, including hypertension, impaired glucose

  2. #2 peak

    - What Are the Issues Concerning Compassion?⏎- Compassion vs Empathy: What Is the Difference?⏎What Is Empathy?⏎Empathy is the ability to put yourself in someone else’s shoes. In other words, it’s the ability

  3. #3 peak

    ?⏎Empathy is either emotional or cognitive. Emotional empathy is the ability to respond with a suitable emotion to another’s mental state. Cognitive empathy is the ability to understand that mental state. These components are independent of each other, which means a

  4. #4 peak

    through oil massage therapy helps rid the body of harmful toxins.<|endoftext|>Object-oriented editing means the ability to edit and/or process individual objects, such as a MIDI or audio clip — whether

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. L3:253 excites
  2. L2:1216 excites
  3. L2:277 excites
  4. L2:410 suppresses
  5. L2:616 suppresses
  6. L3:460 excites
  7. L2:121 excites
  8. L2:488 weakly excites
  9. L3:327 weakly excites
  10. L3:1164 weakly suppresses
  11. L2:62 weakly excites
  12. L2:530 weakly excites
  13. L3:568 weakly excites
  14. L2:102 weakly suppresses
  15. L2:560 weakly excites
  16. L2:1467 weakly suppresses
  17. L0:1017 weakly suppresses
  18. L3:1071 weakly excites
  19. L2:493 weakly suppresses
  20. L3:354 weakly suppresses
  21. L2:2124 weakly excites
  22. L1:2264 weakly excites
  23. L2:327 weakly suppresses
  24. L3:1110 weakly excites

Feeds later layers

  1. L5:2555 suppresses
  2. L12:850 suppresses
  3. L6:1458 suppresses
  4. L7:2264 suppresses
  5. L7:1064 suppresses
  6. L5:1952 suppresses
  7. L9:1444 suppresses
  8. L5:1229 suppresses
  9. L5:853 weakly suppresses
  10. L8:524 weakly suppresses
  11. L6:1428 weakly excites
  12. L7:1540 weakly suppresses
  13. L15:1230 weakly suppresses
  14. L5:692 weakly excites
  15. L9:734 weakly excites
  16. L15:497 weakly excites
  17. L15:989 weakly suppresses
  18. L7:677 weakly suppresses
  19. L7:1560 weakly suppresses
  20. L21:365 weakly excites
  21. L14:1804 weakly excites "and" in subject-area lists
  22. L10:2045 weakly suppresses
  23. L10:2191 weakly suppresses
  24. L12:958 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 -1.77 and -1.581 values between -1.58 and -1.391 values between -1.20 and -1.013 values between -1.01 and -0.8211 values between -0.82 and -0.6417 values between -0.64 and -0.4591 values between -0.45 and -0.26958 values between -0.26 and -0.072259 values between -0.07 and 0.12343 values between 0.12 and 0.31104 values between 0.31 and 0.4941 values between 0.49 and 0.6830 values between 0.68 and 0.8712 values between 0.87 and 1.069 values between 1.06 and 1.257 values between 1.25 and 1.434 values between 1.43 and 1.624 values between 1.62 and 1.811 values between 1.81 and 2.002 values between 2.00 and 2.191 values between 2.19 and 2.383 values between 2.38 and 2.561 values between 2.56 and 2.751 values between 2.75 and 2.941 values between 3.50 and 3.691 values between 3.69 and 3.881 values between 5.57 and 5.76offnegativepositiveoften

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. 网小编
  2. uation
  3. ぶり
  4. ertools
  5. uate
  6. звуч
  7. abama
  8. uating
  9. laws
  10. пресс

Suppresses

  1. +⏎⏎
  2. steen
  3. ELEMENT
  4. 还有一个
  5. 都没有
  6. 另有
  7. 已经有
  8. 已经有了
  9. +⏎
  10. 什么都不