Neuron Explorer

Qwen3-0.6B

the word 'journal' in study reports

A neuron in layer 20 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 'journal' in news sentences like 'published in the journal'. Turning it up did not produce any clear theme, so its role beyond that word is unclear.

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

    a highly promising new approach for the development of drugs to treat pneumococci.⏎As they report in the journal Angewandte Chemie, the scientists copied the choline architecture of the

  2. #2 peak

    have this channel lack the ability to taste anything sweet, bitter or umami (the flavor of MSG), researchers report today (Mar. 6) in the journal Nature.⏎Taste buds have cells that detect sweet, bitter and umami flavors

  3. #3 peak

    70 percent are either overweight or obese, according to the U.S. Centers for Disease Control and Prevention.⏎Results of the study were published recently in the journal Clinical Chemistry.⏎The U.S. National Heart, Lung, and Blood Institute offers

  4. #4 peak

    the microbe populations in the gut.⏎The study was posted online Dec. 29 in advance of print publication in the journal Proceedings of the National Academy of Sciences.⏎While this is not the first time that

  5. #5 peak

    links (crosslink), ATR and its interacting partner ATRIP become activated and phosphorylated leading to the induction of several downstream pathways. Rabbit Polyclonal to APOL1 The activated ATR/ATRIP complex induces the Fanconi

  6. #6 peak

    Nanoscience and Nanotechnology Institute (NUSNNI), and co-leader of the research team.⏎The findings were reported in the online edition of the journal Nature Communications on 16 March 2016.⏎Electron spin

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 can't (g) (1) (i) (h) (e) (e) (e) (e) (e) (e) (e) (e) (e) (e) (e)

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

    inadequate access to water and sanitation facilities. Nigeria has pledged its commitment to meeting Sustainable Development Goal (SDG) 6, which is to “ensure access to water and sanitation for all” by 2030

  2. #2 peak

    more than just a hunch. The renowned psychologist Abraham Maslow stressed that individuals must have their needs for loving and belonging met before they can grow up to have confidence and self-esteem.⏎According to

  3. #3 peak

    will recommend treatment, if needed.⏎Back to TopReferences⏎Chokroverty S, Avidan AY. Sleep and its disorders. In: Daroff RB, Fenichel GM, Jankovic J, Mazziotta JC,

  4. #4 peak

    Whole System Simulation, Proceedings of the 13th USENIX Security Symposium, 2004.⏎Garfinkel, S. and Shelat, A., "Remembrance of Data Passed: A Study of Disk Sanitization

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. L19:1607 suppresses
  2. L19:1873 excites
  3. L19:3041 excites
  4. L19:761 suppresses
  5. L19:2576 weakly excites
  6. L14:471 weakly suppresses
  7. L18:2361 weakly suppresses
  8. L18:1083 weakly excites
  9. L18:151 weakly suppresses
  10. L19:2611 weakly suppresses
  11. L17:744 weakly suppresses
  12. L19:2562 weakly excites
  13. L18:819 weakly excites
  14. L8:1747 weakly excites
  15. L19:1336 weakly excites
  16. L19:1416 weakly suppresses
  17. L16:1659 weakly excites
  18. L19:1761 weakly excites
  19. L0:1141 weakly excites
  20. L8:1544 weakly excites
  21. L11:589 weakly suppresses
  22. L19:2082 weakly excites
  23. L19:960 weakly excites
  24. L14:2015 weakly suppresses

Feeds later layers

  1. L22:151 suppresses
  2. L21:105 suppresses
  3. L21:12 excites
  4. L23:1913 weakly suppresses
  5. L24:267 weakly excites
  6. L23:1665 weakly suppresses
  7. L21:15 weakly excites
  8. L25:2978 weakly suppresses
  9. L23:206 weakly suppresses
  10. L27:425 weakly excites
  11. L21:24 weakly excites
  12. L23:231 weakly excites
  13. L25:45 weakly excites
  14. L26:287 weakly suppresses
  15. L24:1448 weakly suppresses
  16. L21:215 weakly suppresses
  17. L21:17 weakly excites
  18. L27:919 weakly suppresses
  19. L25:108 weakly excites
  20. L27:1664 weakly suppresses
  21. L27:1873 weakly suppresses
  22. L21:16 weakly excites
  23. L22:1078 weakly suppresses
  24. L21:597 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 -11.45 and -10.991 values between -6.46 and -6.011 values between -5.10 and -4.652 values between -2.83 and -2.387 values between -1.93 and -1.479 values between -1.47 and -1.0230 values between -1.02 and -0.57681 values between -0.57 and -0.122554 values between -0.12 and 0.34575 values between 0.34 and 0.7934 values between 0.79 and 1.255 values between 1.25 and 1.702 values between 2.15 and 2.611 values between 3.06 and 3.512 values between 4.87 and 5.321 values between 5.32 and 5.781 values between 5.78 and 6.231 values between 6.23 and 6.68offnegativepositiveoften

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. sup
  2. ypy
  3. YP
  4. trial
  5. .labelX
  6. ular
  7. ahkan
  8. ers
  9. ified
  10. Trial

Suppresses

  1. _POLICY
  2. stituição
  3. otre
  4. 动工
  5. -sales
  6. mutable
  7. anness
  8. zeń
  9. mantle
  10. Birliği