Neuron Explorer

Qwen3-0.6B

"in" after a publication or study word

A neuron in layer 26 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 lights up on the word "in" when it comes right after a study, article or publication, as in "published in" or "article in" a journal. When it was turned up, the model's writing did not become noticeably related to journals or studies.

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

    figure wasn’t statistically significant.⏎“Even total mortality went in the right direction,” said Gaziano, whose new data comes from the Physicians Health Study II published Wednesday in the Journal of the American

  2. #2 peak

    fetal genetic testing extend beyond the health of a baby to screen for desirable physical and mental traits?⏎A recent study conducted by the Georgetown University Medical Center in Washington published in Genetics in Medicine, found that 70 percent of health care practitioners surveyed had

  3. #3 peak

    the current recommendation for five years, say researchers.⏎University of Oxford scientists say the doubling cuts both the risk of the cancer returning and deaths from the disease.⏎Writing in The Lancet, they say this outweighs the risk of side effects.⏎C

  4. #4 peak

    children under 10 years of age and women had a lower incidence of infection than adolescents or adults and men.⏎As the researchers wrote in their study on Tuesday in the New England Journal of Medicine, COVID-19 was first diagnosed in Iceland

  5. #5 peak

    .⏎- Excellent cutting surface roughness can be produced even when machining at low speed.⏎- Suitable for soft materials or workpieces<|endoftext|>Fernanda vertelt haar verhaal in een gesprek met CSL Behring⏎Living With Alpha

  6. #6 peak

    53 days, using a carefully selected array of instruments to peer deeper into the planet than ever before.⏎The most recent findings from these measurements have now been published in a series of papers, revealing the three-dimensional structure of Jupiter’s weather systems—including

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: Justimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimatelyimately

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

    λόφῳ in Einklang zu bringen, ist es unumgänglich nothwendig dass der Ausdruck Roma quadrata technisch in doppeltem Sinne gebraucht wurde; einmal zur Be

  2. #2 peak

    tasks for small-, mini- and micro-satellites.<|endoftext|>Until 1864, Georgia had been barely touched by the war. The Confederate victory at Chickamauga in 1863 had seen off a previous Union invasion

  3. #3 peak

    control of the Chesapeake Bay from the British, victory by the American and French armies at Yorktown would have been impossible. Without the French victory at the Battle of the Capes, American independence from Great Britain might never have been achieved.⏎With

  4. #4 peak

    several times and Washington stopped over at the Hartsville Inn on several occasions, traveling between Philadelphia and Coryell's Ferry (New Hope). The action of the Battle of Crooked Billet took place just south of the Borough; the territory of what

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. L25:247 excites
  2. L20:593 excites
  3. L18:2153 weakly suppresses
  4. L20:64 weakly suppresses
  5. L21:310 weakly excites
  6. L19:948 weakly suppresses
  7. L13:2416 weakly excites
  8. L22:171 weakly excites
  9. L21:852 weakly excites
  10. L23:373 weakly excites 'to' after a noun or adjective
  11. L18:861 weakly excites
  12. L23:128 weakly excites 'for' as in 'for example'
  13. L4:4 weakly suppresses
  14. L19:1305 weakly suppresses unclear: verbs after a noun or 'that'
  15. L22:353 weakly suppresses
  16. L8:328 weakly suppresses
  17. L20:219 weakly suppresses "the" after time or place prepositions
  18. L23:1802 weakly suppresses
  19. L6:388 weakly excites
  20. L18:131 weakly suppresses
  21. L15:267 weakly excites
  22. L3:962 weakly suppresses
  23. L15:1304 weakly suppresses
  24. L22:2991 weakly suppresses

Feeds later layers

  1. L27:647 strongly excites
  2. L27:994 weakly excites
  3. L27:149 weakly excites
  4. L27:854 weakly excites
  5. L27:3052 weakly suppresses
  6. L27:2770 weakly excites
  7. L27:734 weakly excites
  8. L27:2731 weakly excites
  9. L27:108 weakly excites
  10. L27:2655 weakly excites
  11. L27:1357 weakly excites
  12. L27:2876 weakly excites
  13. L27:2204 weakly excites
  14. L27:2365 weakly suppresses
  15. L27:2404 weakly suppresses
  16. L27:2779 weakly suppresses
  17. L27:59 weakly excites full stops in running text
  18. L27:1358 weakly suppresses
  19. L27:2034 weakly excites
  20. L27:1856 weakly excites
  21. L27:2962 weakly excites
  22. L27:2957 weakly suppresses
  23. L27:1325 weakly suppresses
  24. L27:1638 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 -32.25 and -30.651 values between -13.00 and -11.393 values between -11.39 and -9.799 values between -9.79 and -8.196 values between -8.19 and -6.5819 values between -6.58 and -4.9835 values between -4.98 and -3.37102 values between -3.37 and -1.77916 values between -1.77 and -0.162319 values between -0.16 and 1.44232 values between 1.44 and 3.05108 values between 3.05 and 4.6573 values between 4.65 and 6.2529 values between 6.25 and 7.8626 values between 7.86 and 9.469 values between 9.46 and 11.078 values between 11.07 and 12.672 values between 12.67 and 14.283 values between 14.28 and 15.881 values between 17.48 and 19.091 values between 19.09 and 20.692 values between 22.30 and 23.901 values between 23.90 and 25.511 values between 27.11 and 28.711 values between 30.32 and 31.92offnegativepositiveoften

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. The
  2. The
  3. >The
  4. →The
  5. nThe
  6. .The
  7. THE
  8. (The
  9. "The
  10. :The

Suppresses

  1. imately
  2. toMatch
  3. みると
  4. inness
  5. EntityState
  6. #endregion
  7. der
  8. ⏎ ⏎
  9. webElementProperties
  10. ymoon