Neuron Explorer

Qwen3-0.6B

unclear: mid-sentence words before a verb

A neuron in layer 22 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 gives weak, scattered responses to words in the middle of a sentence, like 'can', 'just' or 'easily', often just before an action verb. There is no clear theme in its strongest examples. Turning it up or down changed the text in ways that did not seem connected to these words.

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

    are especially likely to have severe problems with RSV.⏎There are several ways to test for RSV. We can grow the virus in cultured cells, or we can look for antibodies to the virus in a patient's blood (although we have to

  2. #2 peak

    a little broom, which makes this sport unlike any other.⏎What is a bonspiel?⏎A bonspiel is a curling tournament.⏎Why don’t they just call it a curling tournament?⏎“Bonspiel” better captures the specialness

  3. #3 peak

    in a Venetian gondola. I also owe a deep professional debt to Michael Colacurcio for years of guidance, as well as to Eric Sundquist ...⏎Introduction: Education and the "Alexandrian Library"⏎An Adver

  4. #4 peak

    being its creator.⏎Cavorite has a greater presence in the comic book world (including a 'Doctor Who' story!), but that is of no use to Toobworld......

  5. #5 peak

    guess what order they will spot them in!⏎✍Make up the checklist together – we’ve used stickers from our Out and About sticker pack but you can just as easily draw or write on some blank stickers.⏎👀 On your walk, encourage your child

  6. #6 peak

    with a “Spanish” hat made of velvet.⏎While it is unclear who exactly designed the crown, credit usually goes to the chief designer at Rundells, Philip Liebart. The design incorporates symbols of the three countries of the

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 need to be able to be the most to be the most to be the most to be the most to be the most to be the most to be the most to be the most to be the most to be the most to be 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

    66 the Fisk expedition explored the gold fields of Montana. Illingworth accompanied the expedition as a photographer, and, in spite of the technical challenges this expedition presented for the execution of wetplate negatives, Illingworth managed to create about

  2. #2 peak

    MA, its name in Spanish), was inaugurated to highlight the voices of women survivors of the clandestine centres, shedding light on the acts of gender-based violence they underwent during their detention, and the impact that it had in their lives after the

  3. #3 peak

    adholder of Holland, who became King of England but remained passionately Dutch. This article commemorates the 300th anniversary of his death⏎The article you want to access is behind a paywall. You can purchase this article or subscribe

  4. #4 peak

    roots in different cultures around the world through to the present day. He explores crucial milestones in scientific development and at the same time examines the enormous social and intellectual changes they initiated. Opening with a discussion of the key elements of modern scientific enterprise, the

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. L20:1727 excites
  2. L19:385 suppresses
  3. L20:1124 suppresses
  4. L20:1962 excites
  5. L21:1646 excites
  6. L21:1416 excites
  7. L20:2482 suppresses
  8. L21:935 excites
  9. L21:914 excites
  10. L19:504 suppresses
  11. L19:1769 excites
  12. L21:600 suppresses
  13. L19:1166 suppresses
  14. L21:200 suppresses
  15. L20:585 excites
  16. L20:1652 excites
  17. L16:1153 suppresses
  18. L19:2567 excites
  19. L19:2632 suppresses
  20. L21:487 excites
  21. L14:659 excites
  22. L19:1622 suppresses
  23. L21:1569 excites
  24. L21:594 excites

Feeds later layers

  1. L23:1060 excites
  2. L24:2215 excites
  3. L24:1263 weakly excites
  4. L24:393 weakly excites
  5. L23:2179 weakly suppresses
  6. L24:57 weakly suppresses
  7. L27:1487 weakly suppresses
  8. L27:756 weakly excites
  9. L25:1161 weakly suppresses
  10. L24:293 weakly suppresses
  11. L23:1235 weakly excites
  12. L25:1641 weakly suppresses
  13. L25:1311 weakly excites
  14. L27:1184 weakly suppresses
  15. L24:565 weakly excites
  16. L26:3055 weakly excites
  17. L25:2661 weakly excites
  18. L25:1849 weakly suppresses
  19. L23:1815 weakly excites
  20. L25:1342 weakly suppresses
  21. L26:1225 weakly excites
  22. L23:271 weakly suppresses
  23. L24:1200 weakly suppresses
  24. L24:302 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 -9.77 and -9.371 values between -8.97 and -8.572 values between -8.57 and -8.172 values between -6.97 and -6.574 values between -6.57 and -6.171 values between -6.17 and -5.765 values between -5.76 and -5.364 values between -5.36 and -4.967 values between -4.96 and -4.569 values between -4.56 and -4.1612 values between -4.16 and -3.7612 values between -3.76 and -3.3618 values between -3.36 and -2.9620 values between -2.96 and -2.5619 values between -2.56 and -2.1620 values between -2.16 and -1.7548 values between -1.75 and -1.3569 values between -1.35 and -0.95126 values between -0.95 and -0.55489 values between -0.55 and -0.151843 values between -0.15 and 0.25950 values between 0.25 and 0.65192 values between 0.65 and 1.0521 values between 1.05 and 1.458 values between 1.45 and 1.869 values between 1.86 and 2.267 values between 2.26 and 2.665 values between 2.66 and 3.062 values between 3.06 and 3.461 values between 4.66 and 5.061 values between 5.87 and 6.27offnegativepositiveoften

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. put
  2. 做
  3. done
  4. 有
  5. 来做
  6. Put
  7. 去做
  8. 说
  9. doing
  10. go

Suppresses

  1. cly
  2. 实施
  3. roker
  4. reu
  5. ifiant
  6. 执行
  7. executing
  8. 遵循
  9. eer
  10. engl