Neuron Explorer

Qwen3-0.6B

'to' after a verb, before 'avoid'

A neuron in layer 25 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 word 'to' when it introduces a purpose, as in 'to avoid cross-contamination' or 'to reduce'. Turning it down made the text repeat the word 'avoid' over and over, which matches what it is about.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    causes baking soda to foam up. If stored in a closed container, the mixture can explode.⏎What should you not mix with vinegar?⏎3 Cleaning Products That You Should Not Mix With Vinegar⏎- Bleach And Vinegar. Bleach and

  2. #2 peak

    ained soil.⏎- The seedlings should be placed 8 to 18 inches apart, depending on the variety. Ward off any flower while sowing, to redirect the plants' energy to the roots. This will ensure long-term health and

  3. #3 peak

    liquidity. Without safeguards for liquidity in place, a "bank run" can incur, which any banking system wants to avoid at any cost.⏎Why it Matters:⏎Any entity which needs to generate and

  4. #4 peak

    chicken. For example, if you are baking six breast pieces, each one might cook differently based on size. Don’t forget to clean your thermometer probe between readings to avoid cross-contamination.⏎Food or meat thermometers are fairly inexpensive, and will

  5. #5 peak

    by:⏎- Cleaning your hands frequently using either soap and warm water or alcohol-based hand sanitizers,⏎- Covering your cough and sneezing into your arm to reduce

  6. #6 peak

    quickly than we do when they feel danger.⏎Your horse may see that white bear, shaped like a bag, long before you do. So, he jumps sideways to get away from it. More than likely, he doesn’t think there is a

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: avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid avoid

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

    sticky substance known as honeydew. This, in turn, attracts sooty mold, which grows wherever the honeydew is present. If the mold infestation is heavy, it may block photosynthesis and kill the

  2. #2 peak

    have a crush on someone: I used to have a massive crush on my geography teacher.⏎Is a crush love?⏎Crush is defined as a brief but intense infatuation for someone, especially someone inappropriate or unattainable. … Unlike

  3. #3 peak

    , often asymmetric at the base, abruptly sharp-pointed at the tip, saw-toothed on the margins, and usually somewhat hairy beneath.⏎Flowers: Inflorescence of numerous, small flowers in a 4-10 cm

  4. #4 peak

    which is relatively sparsely populated, before expanding elsewhere.) But wait, I ask him, not wanting to get taken in by the techno-optimism that infuses so much of what Google does these days: what are the chances this will

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. L22:1806 strongly excites "to" introducing a purpose clause
  2. L22:2315 excites
  3. L22:3032 suppresses
  4. L21:1397 suppresses
  5. L19:2282 suppresses
  6. L23:872 suppresses
  7. L23:1503 suppresses
  8. L23:434 excites
  9. L22:1403 weakly suppresses
  10. L15:2177 weakly suppresses
  11. L9:146 weakly excites
  12. L2:75 weakly suppresses
  13. L22:530 weakly suppresses
  14. L20:2898 weakly suppresses
  15. L20:1018 weakly excites
  16. L18:3059 weakly suppresses
  17. L19:1343 weakly suppresses
  18. L24:385 weakly excites
  19. L21:2461 weakly suppresses
  20. L17:1187 weakly suppresses
  21. L1:2277 weakly excites
  22. L3:2078 weakly excites
  23. L21:1678 weakly suppresses
  24. L19:347 weakly excites

Feeds later layers

  1. L27:2185 suppresses
  2. L26:3012 weakly suppresses
  3. L26:2381 weakly suppresses
  4. L27:1591 weakly suppresses
  5. L26:408 weakly excites
  6. L26:438 weakly suppresses
  7. L26:1997 weakly suppresses
  8. L26:2662 weakly suppresses
  9. L27:2814 weakly suppresses
  10. L26:271 weakly suppresses
  11. L26:1055 weakly excites
  12. L27:1279 weakly suppresses
  13. L26:406 weakly suppresses
  14. L26:1454 weakly suppresses
  15. L27:1448 weakly excites
  16. L27:2800 weakly suppresses
  17. L27:790 weakly excites
  18. L27:203 weakly excites
  19. L26:2752 weakly excites
  20. L27:1137 weakly suppresses
  21. L27:2341 weakly suppresses
  22. L27:750 weakly excites
  23. L27:686 weakly excites
  24. L27:398 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 -14.37 and -13.181 values between -9.62 and -8.432 values between -6.06 and -4.872 values between -4.87 and -3.684 values between -3.68 and -2.4911 values between -2.49 and -1.31993 values between -1.31 and -0.122799 values between -0.12 and 1.0748 values between 1.07 and 2.2615 values between 2.26 and 3.4417 values between 3.44 and 4.634 values between 4.63 and 5.821 values between 5.82 and 7.002 values between 7.00 and 8.193 values between 8.19 and 9.382 values between 10.57 and 11.751 values between 12.94 and 14.131 values between 30.75 and 31.941 values between 31.94 and 33.13offnegativepositiveoften

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. всего
  2. NameValuePair
  3. ところで
  4. TestCase
  5. pez
  6. INA
  7. roe
  8. _tokenize
  9. ALK
  10. (criteria

Suppresses

  1. 避免
  2. avoid
  3. avoid
  4. avoided
  5. avoids
  6. avoiding
  7. Avoid
  8. Avoid
  9. avoidance
  10. 防止