Neuron Explorer

Qwen3-0.6B

word starts like cr, fr, kr

A neuron in layer 26 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the first consonant pieces of words split into tokens, such as "cr" in craved, "fr" in frigid, and "Kr" in Kraken. It goes strongly negative at the end of a quoted phrase, such as "galaxies" in quotes or "reality" in a quote. Turning it up made the text fill with quotation marks, which fits the quote marks it pushes toward in the output.

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

    money from a broker to make a trade, which is allowed on many exchanges and brokerages. Margin trading is available on many cryptocurrency exchanges at this point, including Kraken and Binance.⏎- Binary Options Trading⏎It is possible to trade binary

  2. #2 peak

    . Practice donating.⏎Have your children practice donating something to a family in need<|endoftext|>Severely corroded pipelines will force BP to shut down its oil field at Prudhoe Bay, Alaska, company officials said on Sunday. After a major spill

  3. #3 peak

    the word mostly on my mind. I was self-isolating for two weeks which game me plenty of time to think. I enjoyed my solitude but I also craved connections with my family and friends. The Psychologies Life Leap Club set up

  4. #4 peak

    century.⏎When astronauts discover the huge expanse of outer house, they should put on high-tech spacesuits to guard them from the frigid situations of the cosmos.⏎Hollywood motion pictures have glamorized the long

  5. #5 peak

    Myth: Good swimmers don’t drown.⏎Reality: Anyone can drown. Even the best swimmers, can tire, drift too far out, get a cramp, suffer an injury, or get caught in a current.⏎Myth:

  6. #6 peak

    underwater. They can be warm in frigid weather; they can travel and move across deserts. People can live anywhere

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:'s' seem' seem't's's's's's's's's's's's's's's's's's's's't's's's's't's's's's't's's's't's's's't's's't's

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

    suspected galaxies could be before," said study team member Daniel Zucker of Cambridge University. "So rather than dwarf galaxies, we should perhaps call them 'hobbit galaxies.'"⏎The dimness could be the result of stellar age, as seven of

  2. #2 peak

    for one to find further examples on one's own. In short, metafiction happens when the author inserts anything that "breaks up the illusion of 'reality' in a work" (Dr. Wheeler, "Literary Terms and Definitions

  3. #3 peak

    ten years, this is the challenge," Barber says of efforts to bring such natural designs into engineering. (See "Twisting Everyday Fibers Could Make 'Smart Clothes' a Reality.")⏎Performance engineers drawing inspiration from a limpet? That's<|endoftext|>

  4. #4 peak

    Jack, echoing the opinion of the other 15 boys at the experimental Aspie school here in the Catskills. ''So we can't be 'cured.' This is just the way we are.''⏎From behind his GameBoy,

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. L23:525 strongly excites
  2. L25:278 strongly suppresses
  3. L24:1725 strongly excites
  4. L24:1505 strongly excites
  5. L24:1199 strongly suppresses
  6. L20:1462 excites
  7. L25:470 suppresses
  8. L24:74 suppresses
  9. L22:397 suppresses
  10. L25:235 suppresses
  11. L24:1414 excites
  12. L20:2332 excites
  13. L22:1740 excites
  14. L22:768 suppresses
  15. L24:286 suppresses
  16. L24:874 suppresses
  17. L24:701 excites
  18. L21:715 excites
  19. L25:946 suppresses
  20. L21:988 excites
  21. L19:129 suppresses
  22. L22:1668 excites
  23. L23:293 suppresses
  24. L21:679 excites

Feeds later layers

  1. L27:1108 strongly excites
  2. L27:865 strongly excites
  3. L27:1294 strongly suppresses
  4. L27:444 excites
  5. L27:277 excites
  6. L27:986 excites
  7. L27:2161 excites
  8. L27:661 weakly excites
  9. L27:499 weakly excites
  10. L27:905 weakly suppresses
  11. L27:402 weakly excites
  12. L27:806 weakly excites
  13. L27:1955 weakly suppresses
  14. L27:1064 weakly suppresses
  15. L27:2714 weakly excites
  16. L27:904 weakly excites
  17. L27:1370 weakly excites
  18. L27:2800 weakly suppresses
  19. L27:1839 weakly suppresses
  20. L27:997 weakly suppresses
  21. L27:1712 weakly excites
  22. L27:341 weakly excites
  23. L27:2601 weakly suppresses
  24. L27:235 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 -31.52 and -30.381 values between -21.32 and -20.181 values between -20.18 and -19.052 values between -9.99 and -8.851 values between -8.85 and -7.722 values between -7.72 and -6.593 values between -6.59 and -5.452 values between -5.45 and -4.323 values between -4.32 and -3.196 values between -3.19 and -2.0557 values between -2.05 and -0.923469 values between -0.92 and 0.21318 values between 0.21 and 1.3518 values between 1.35 and 2.486 values between 2.48 and 3.615 values between 3.61 and 4.754 values between 4.75 and 5.883 values between 5.88 and 7.011 values between 7.01 and 8.152 values between 8.15 and 9.281 values between 9.28 and 10.411 values between 11.55 and 12.681 values between 12.68 and 13.81offnegativepositiveoften

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. '
  2. 」
  3. ‘
  4. 」⏎⏎
  5. ‘
  6. "'⏎
  7. "',
  8. »
  9. "',⏎
  10. ",'

Suppresses

  1. ']}'
  2. dorf
  3. avored
  4. '")⏎
  5. 镇政府
  6. andoned
  7. orean
  8. <?,
  9. 's
  10. 们的