Neuron Explorer

Qwen3-0.6B

'that' after a noun, 'being' in passives

A neuron in layer 15 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 responds in two ways: it fires on the word 'that' right after a noun that starts a describing clause, like 'the financial issues that break us'. It goes strongly negative on 'being' in phrases like 'the behavior being pressured'. Turning it up or down gave mostly unrelated, looping 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

    from Kiku Hughes.⏎Kiku is on vacation in San Francisco when suddenly she finds herself displaced to the 1940s Japanese-American internment camp that her late grandmother, Ernestina, was forcibly relocated to during World War II.⏎

  2. #2 peak

    ability to speak a first language, are thought to be due to the inbuilt genetic process. It is said that children naturally become interested in the sounds and signs that they see around them and the ability to speak ‘unfolds’.⏎Although

  3. #3 peak

    can easily incorporate this kind of activity into your daily regimen. Simple activities for preschoolers must enable them to discover about shades, shapes, as well as other items that interest them.⏎A deck of playing cards can be reworked right into a number

  4. #4 peak

    then.) He had just gotten his PhD at Caltech under Richard Feynman, and was busy trying to understand the “zoo” of hadronic particles that were being discovered in the 1960’s. (Side note:

  5. #5 peak

    This teaches your tween about the importance of communicating about money.⏎#5. Talk About Financial Options.⏎One of the hardest parts of life is encountering the financial issues that break us. Talking about the different

  6. #6 peak

    Weingarten: Most of our educational systems have been created to an industrial model. They were built to ensure that kids could competently deal with the routine tasks that were necessary in factories. Some kids would succeed in a different way; but in

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: I'm writing the problem I'm thinking about, and I'm trying to consider the problem I'm writing about I'm referring to, that I'm referring to? I'm Thinking of (I'm Thinking of) That I'm Writing?

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

    child attempting to reproduce them based on a child’s own free will. Peer pressure is the act of one child coercing another to follow suit. Often the behavior being pressured is questionable or taboo, such as smoking cigarettes or drinking alcohol.⏎Peer Pressure

  2. #2 peak

    the decimal and cents to the right of the decimal. To add decimals, simply line up the decimal points of the amounts of money that are being added. When you have added the amounts of money, be sure to place the

  3. #3 peak

    language around 1600, the word meant to cook something slowly and gently in water that is just below the boiling point. Usually eggs are what's being cooked when "coddle" is used in this sense today, but when

  4. #4 peak

    More recently, a myopic focus on growth has created a dream that promotes unbridled consumption, but the “more is better” version actually harms the very things we cherish. It is unsustainable personally, as it draws American families into a work

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. L14:2528 suppresses
  2. L14:2055 suppresses preposition after a noun phrase
  3. L14:492 suppresses
  4. L7:2503 suppresses
  5. L9:2312 excites 'to' after a noun (a role to play)
  6. L12:1667 excites
  7. L13:1234 suppresses
  8. L12:2971 suppresses
  9. L12:2722 suppresses
  10. L13:1785 suppresses content nouns in health and advice text
  11. L4:73 suppresses
  12. L4:470 weakly suppresses
  13. L13:1173 weakly excites
  14. L10:1162 weakly excites
  15. L14:2666 weakly suppresses
  16. L10:1256 weakly excites
  17. L11:1505 weakly excites
  18. L6:958 weakly suppresses
  19. L14:1849 weakly excites
  20. L9:865 weakly suppresses
  21. L10:353 weakly excites
  22. L14:2265 weakly excites
  23. L13:2373 weakly suppresses
  24. L7:2751 weakly suppresses

Feeds later layers

  1. L19:2179 suppresses
  2. L16:1381 suppresses
  3. L18:1672 suppresses
  4. L16:2806 suppresses
  5. L21:2602 suppresses
  6. L20:2132 suppresses "the" starting an object noun phrase
  7. L19:1235 suppresses
  8. L22:2064 suppresses
  9. L17:2783 suppresses
  10. L18:2378 suppresses
  11. L17:2557 suppresses
  12. L16:550 suppresses
  13. L19:1483 suppresses
  14. L18:1342 suppresses
  15. L21:2465 suppresses
  16. L21:1422 suppresses
  17. L16:2494 weakly suppresses
  18. L17:1884 weakly suppresses
  19. L20:1324 weakly suppresses
  20. L19:2621 weakly suppresses
  21. L17:1918 weakly suppresses
  22. L21:2163 weakly suppresses
  23. L18:1520 weakly suppresses
  24. L17:451 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 -7.30 and -6.941 values between -4.09 and -3.731 values between -3.38 and -3.021 values between -2.66 and -2.301 values between -2.30 and -1.951 values between -1.95 and -1.593 values between -1.59 and -1.234 values between -1.23 and -0.8857 values between -0.88 and -0.521106 values between -0.52 and -0.162467 values between -0.16 and 0.20193 values between 0.20 and 0.5538 values between 0.55 and 0.9112 values between 0.91 and 1.2710 values between 1.27 and 1.623 values between 1.62 and 1.982 values between 1.98 and 2.343 values between 2.34 and 2.692 values between 2.69 and 3.051 values between 4.12 and 4.481 values between 6.62 and 6.98offnegativepositiveoften

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. alley
  2. estyle
  3. assium
  4. ucer
  5. образом
  6. 裾
  7. .initState
  8. egal
  9. ande
  10. unday

Suppresses

  1. Boards
  2. ICATION
  3. 所
  4. whom
  5. 暄
  6. ronic
  7. 所需要
  8. 所带来的
  9. contemplated
  10. 所致