Neuron Explorer

Qwen3-0.6B

'hands' in hand-washing advice

A neuron in layer 11 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 'hands' in health advice about washing them, like 'wash your hands' during illness. The strongest negative cases are unrelated product or spec wording, so the negative side is not very meaningful. Turning it up did not bring out hand-washing text.

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

    support fomite transmission. No evidence to say that washing your hands, using sanitiser gel, wiping down surfaces, etc will make you safer.

  2. #2 peak

    the person's vaccine history. A booster shot may be given. Antibiotics may also be given, even if the person does not have symptoms.⏎- Wash your hands often. Wash your hands<|endoftext|>Online, Maria Droujkova has founded the

  3. #3 peak

    failures on basic sanitation levels. In the old model the water tanks were often insufficient, which meant that nurses had to see consecutive patients without washing their hands.⏎Other problems include an inability to deal with the extreme summer (and winter)

  4. #4 peak

    a week, even if it's just a short walk or online class.⏎Get as much sleep as you can, including naps if possible.⏎Wash your hands often and avoid touching your eyes to avoid catching a virus, especially if your baby

  5. #5 peak

    in just a couple of days. It is important to wash hands often and avert close contact.⏎- Subconjunctival hemorrhage. An

  6. #6 peak

    steps we can all take to help assure our businesses remain open and that we prevent the further spread of this awful disease in our communities. Everyone must still wash their hands, not touch their faces, stay home if their sick, cough into a tissue

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 most efficient way to create a system that can be used for both types of data (a) or (b), (a) or (b) (a) or (b) (a) or (b) (

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

    product categories.⏎If the one list has to be by category then colour or pack size and all the products use the same content type, you could use one view to sort by category then colour then pack size. There will be no overlap or conflict

  2. #2 peak

    management. Aqua Gen and Namdal Rensefisk have stringent requirements to filtering and temperature control, which greatly impact the facility's complexity. Instead of placing a single ring pipe under all of the tanks and basic outlet water pipes from the tanks and

  3. #3 peak

    prevent corrosion, though stainless steel will provide greater corrosion-resistance. Zinc plating is usually used for aesthetic purposes.⏎All threaded rod in this selection are available in one meter lengths.⏎Custom Cutting and Chamfer Options Available. Call For Details⏎What

  4. #4 peak

    alerts. The global financial system—bank transfers, stock trades—increasingly relies on it.⏎Because of abstraction and the distributed nature of cloud computing, no single data centre holds all the world’s valuables. No single cable keeps the entire

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. L8:972 excites
  2. L5:1119 excites
  3. L4:961 excites
  4. L10:2308 excites
  5. L7:2007 weakly excites
  6. L10:1284 weakly excites
  7. L8:2882 weakly suppresses
  8. L10:1080 weakly excites weight loss and body fat words
  9. L4:2027 weakly suppresses
  10. L7:2357 weakly suppresses
  11. L8:2065 weakly suppresses
  12. L0:872 weakly suppresses
  13. L1:1898 weakly excites
  14. L5:559 weakly suppresses
  15. L10:828 weakly suppresses
  16. L9:2305 weakly suppresses
  17. L10:916 weakly excites
  18. L8:2220 weakly suppresses
  19. L4:644 weakly excites
  20. L8:2489 weakly excites
  21. L6:2216 weakly suppresses
  22. L1:942 weakly excites
  23. L9:2761 weakly suppresses
  24. L9:2172 weakly excites

Feeds later layers

  1. L12:2717 suppresses
  2. L14:2870 suppresses
  3. L13:2629 excites
  4. L20:1148 suppresses
  5. L18:2955 suppresses
  6. L25:1187 suppresses
  7. L13:913 suppresses
  8. L22:1176 suppresses
  9. L15:2275 suppresses
  10. L15:1717 suppresses
  11. L14:2783 suppresses the word 'same' in 'the same'
  12. L13:2117 suppresses
  13. L13:919 suppresses
  14. L13:1586 weakly suppresses
  15. L17:2535 weakly suppresses
  16. L14:1990 weakly suppresses
  17. L12:225 weakly suppresses
  18. L14:1669 weakly suppresses
  19. L19:2233 weakly suppresses
  20. L18:833 weakly suppresses
  21. L13:1266 weakly suppresses
  22. L20:1862 weakly suppresses
  23. L14:2829 weakly suppresses
  24. L19:2118 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 -5.30 and -5.111 values between -4.00 and -3.821 values between -2.70 and -2.521 values between -1.96 and -1.781 values between -1.78 and -1.591 values between -1.59 and -1.415 values between -1.22 and -1.044 values between -0.85 and -0.6712 values between -0.67 and -0.4887 values between -0.48 and -0.30599 values between -0.30 and -0.112109 values between -0.11 and 0.08933 values between 0.08 and 0.26134 values between 0.26 and 0.4515 values between 0.45 and 0.631 values between 0.63 and 0.821 values between 1.00 and 1.191 values between 1.19 and 1.371 values between 1.93 and 2.11offnegativepositiveoften

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. 현
  2. detalles
  3. détail
  4. enser
  5. 原型
  6. 细节
  7. urre
  8. delaying
  9. ұ
  10. iciary

Suppresses

  1. universal
  2. Universal
  3. 就够了
  4. 统一
  5. universal
  6. ��
  7. unified
  8. 一站式
  9. 全局
  10. 一口