Neuron Explorer

Qwen3-0.6B

hyphen inside COVID-19 and similar compounds

A neuron in layer 0 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 reacts to the dash in the word COVID-19, and to dashes that split other compound words such as mis-information. When it was turned up, the model's replies did not visibly change in a way that relates to this.

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

    yourself from COVID-19 is by using an alcohol-based massage or washing your hands frequently, and not touching your mouth, nose, and eyes.⏎The COVID-19 virus spreads mainly through dewdrops of saliva or discharge from the nose when

  2. #2 peak

    work. Women and children are especially vulnerable to violence both within and outside their homes. Substance abuse is high and accessing affordable healthcare is a huge challenge.⏎The COVID-19 pandemic has worsened their living conditions and led to loss of income,

  3. #3 peak

    coronavirus pandemic we're currently experiencing. It swept across the world fast and continues to ravish every country on the face of the earth.⏎The explosive spread of COVID-19 lead to mass panic. Essential supplies like food and toilet paper flew off

  4. #4 peak

    all-time high. Another concerning development was the slowdown in fuel economy improvements around the world as car buyers continued to purchase bigger vehicles.⏎The report adds that the COVID-19 crisis is disrupting important progress that has been achieved in key areas in recent

  5. #5 peak

    until the end of 2024.⏎The primary purpose behind implementing this law is to ensure that the people of certain industries who got affected because of COVID-19 can come back and work again. This only applies to the employees that

  6. #6 peak

    2020, shows that even before the COVID-19 pandemic, majorities of citizens reported going without needed medical care and clean

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 value of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the number of the

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

    These trends are concerning. Recent research has demonstrated that people are very poor at identifying manipulated images. And while the technology companies have vowed to combat the sharing of mis- and dis-information, their efforts have focused solely on

  2. #2 peak

    . Sacramento, CA.⏎- Quinn, L., B. Davis, et al. (2006). Does horse manure harbor invasive plants? Cal-IPC Symposium. Rohnert Park,

  3. #3 peak

    , and a letter from John Ridge to Georgia governor Wilson Lumpkin dated September 22, 1833 notifies the governor of murders committed against pro-removal Cherokees by anti-removal Cherokees.<|endoftext|>Create

  4. #4 peak

    -IPC Inventory rating: Moderate⏎Cal-IPC Resources on Rumex acetosella⏎- California Invasive Plant Inventory Plant Assessment Form - Information gathered by Cal-IPC on the impacts, rate of spread, and distribution of invasive plants in California

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

None: this is the first layer.

Feeds later layers

  1. L23:2786 weakly excites
  2. L11:973 weakly suppresses
  3. L24:1834 weakly excites
  4. L15:2184 weakly suppresses
  5. L10:1549 weakly excites
  6. L12:1111 weakly suppresses
  7. L13:2019 weakly excites
  8. L14:371 weakly suppresses
  9. L14:2966 weakly excites
  10. L15:2658 weakly excites
  11. L9:1051 weakly suppresses
  12. L1:2339 weakly suppresses
  13. L8:911 weakly excites
  14. L8:760 weakly suppresses
  15. L2:812 weakly suppresses
  16. L17:2873 weakly excites
  17. L16:1449 weakly suppresses
  18. L22:2199 weakly excites
  19. L6:1161 weakly suppresses
  20. L23:2140 weakly excites
  21. L6:1510 weakly suppresses
  22. L15:1675 weakly excites
  23. L8:866 weakly excites
  24. L8:1640 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 -1.96 and -1.906 values between -0.27 and -0.2125 values between -0.21 and -0.15142 values between -0.15 and -0.09553 values between -0.09 and -0.031453 values between -0.03 and 0.031137 values between 0.03 and 0.08446 values between 0.08 and 0.14123 values between 0.14 and 0.2015 values between 0.20 and 0.266 values between 0.26 and 0.321 values between 0.32 and 0.38offnegativepositiveoften

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. ado
  2. fert
  3. ão
  4. idan
  5. ake
  6. aptor
  7. 信息发布
  8. ano
  9. ]\\
  10. aos

Suppresses

  1. #elif
  2. ман
  3. _counters
  4. ч
  5. diag
  6. (chars
  7. 痂
  8. &C
  9. #endregion
  10. sched