Neuron Explorer

Qwen3-0.6B

empty slot before a number

A neuron in layer 22 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 spot just before a number, such as in "at age" or "about" followed by a figure like 40 or 25. Turned down, it made the model repeat "has been" and "have", which is what it tends to push against.

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

    process that uses ionized air to increase the surface tension of non-porous substrates. Normally, corona-treating systems operate at an electrical voltage of approximately 10 kV.⏎Like corona, plasma is the electrical ionization of

  2. #2 peak

    a young age).⏎Guidelines differ for men at average risk. Some recommend an initial PSA (Prostate Specific Antigen) blood test and clinical examination at age 40, and others recommend starting at age 50. In general,

  3. #3 peak

    Figure 20-17 Intravitreal injection. The ocular surface was anesthetized with subconjunctival injection of lidocaine, 2%, in the inferotemporal quadrant. An eyelid speculum was

  4. #4 peak

    routine screening tests for people at average risk of cancer. People with higher risk may be screened more often.⏎- Breast cancer: Get yearly mammograms starting at age 40, in addition to regular self-exams and exams by your doctor or

  5. #5 peak

    more susceptible to heat stress – lower-producing cows and Jerseys may be comfortable until about 25°C. High air temperature, humidity, solar radiation, and low air

  6. #6 peak

    , sleep inducing, and muscle spasm relieving effects.⏎The following doses have been studied in scientific research:⏎- For generalized anxiety disorder (GAD):⏎- 45 drops<|endoftext|>BROOKLYN BRIDGE FOREST⏎Our Brooklyn

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: have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have have 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

    such as dandelions and beetle grubs. Scientific evidence shows that these chemicals also affect people, especially children, and pets. Exposure to certain lawn care pesticides has been associated with increased risks of a variety of serious health problems, including asthma,

  2. #2 peak

    Here are some of the key health benefits of broccoli:⏎- May help to reduce the risk of cancer: Broccoli contains sulforaphane, a compound that has been shown to have anti-cancer properties. Studies suggest that sulforaphane

  3. #3 peak

    in diabetes.⏎An apple is low in calories and contains high fiber and for this reason it fills the stomach without increasing blood sugar levels.⏎In some researches, it has been found that apples have anti-diabetic characteristics. People even identified as pre

  4. #4 peak

    college is a great time to form this habit as it will serve you well throughout your life.⏎For one, exercise increases one’s energy levels. It’s also been proven to replenish brain energy reserves, especially in areas of the brain responsible for

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. L18:2009 excites
  2. L17:2758 suppresses
  3. L19:554 suppresses subjects of conditional or advice clauses
  4. L15:2370 excites
  5. L3:1498 weakly suppresses
  6. L12:2355 weakly suppresses
  7. L18:2656 weakly excites
  8. L5:1090 weakly suppresses
  9. L19:2180 weakly suppresses
  10. L0:1111 weakly suppresses
  11. L14:1287 weakly suppresses
  12. L20:2512 weakly suppresses
  13. L20:2256 weakly excites
  14. L20:223 weakly suppresses
  15. L11:509 weakly excites
  16. L18:730 weakly suppresses
  17. L11:3023 weakly suppresses gender and race in lists of identities
  18. L19:2857 weakly suppresses
  19. L9:2700 weakly suppresses
  20. L15:782 weakly suppresses
  21. L9:1329 weakly excites
  22. L20:2894 weakly excites
  23. L9:1010 weakly suppresses
  24. L16:837 weakly suppresses

Feeds later layers

  1. L24:243 strongly excites
  2. L23:1189 strongly suppresses
  3. L26:1914 suppresses
  4. L24:1770 suppresses
  5. L25:613 suppresses
  6. L27:756 suppresses
  7. L23:892 suppresses "the" in "over the past" time spans
  8. L24:60 excites
  9. L25:1726 weakly excites
  10. L23:1727 weakly suppresses
  11. L25:685 weakly excites
  12. L27:1387 weakly suppresses
  13. L24:2257 weakly excites
  14. L26:936 weakly suppresses
  15. L26:848 weakly excites
  16. L23:1164 weakly excites
  17. L23:1013 weakly suppresses
  18. L26:2133 weakly excites
  19. L23:747 weakly suppresses
  20. L25:1567 weakly excites
  21. L25:1995 weakly suppresses
  22. L24:2199 weakly excites
  23. L23:1173 weakly excites
  24. L27:1129 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 -9.84 and -9.421 values between -9.01 and -8.601 values between -8.60 and -8.191 values between -8.19 and -7.782 values between -7.36 and -6.951 values between -6.95 and -6.541 values between -6.54 and -6.131 values between -6.13 and -5.711 values between -5.71 and -5.305 values between -5.30 and -4.891 values between -4.89 and -4.486 values between -4.48 and -4.076 values between -4.07 and -3.657 values between -3.65 and -3.2411 values between -3.24 and -2.8322 values between -2.83 and -2.4222 values between -2.42 and -2.0029 values between -2.00 and -1.5953 values between -1.59 and -1.18107 values between -1.18 and -0.77443 values between -0.77 and -0.361744 values between -0.36 and 0.06925 values between 0.06 and 0.47224 values between 0.47 and 0.88101 values between 0.88 and 1.2954 values between 1.29 and 1.7147 values between 1.71 and 2.1230 values between 2.12 and 2.5323 values between 2.53 and 2.9410 values between 2.94 and 3.3611 values between 3.36 and 3.778 values between 3.77 and 4.181 values between 4.18 and 4.594 values between 4.59 and 5.003 values between 5.00 and 5.421 values between 6.24 and 6.65offnegativepositiveoften

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. anced
  2. ериал
  3. 说不定
  4. ukes
  5. lime
  6. bons
  7. 览
  8. tutors
  9. еб
  10. -mort

Suppresses

  1. 've
  2. ’ve
  3. 已经被
  4. have
  5. 已被
  6. 已经
  7. has
  8. hasn
  9. 已
  10. haven