Neuron Explorer

Qwen3-0.6B

unclear: word pieces in medical terms

A neuron in layer 25 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 is hard to pin down. It fires on word fragments inside long medical or scientific names, such as vitamin E and thyroid terms. When turned up, the model kept writing 'not' and "don't", which hints at a negation role.

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 moved or pressed away from their usual position inside the skull.⏎Herniation syndrome; Transtentorial herniation; Uncal herniation; Subfalcine herniation; Tonsillar herniation; Herniation - brain⏎

  2. #2 peak

    population sizes and genetic diversity at the time of DFTD emergence, owing to elevated devil abundances following the extinction of the previous apex predator - the thylacine - approximately 80. years ago.<|endoftext|>Drupal is a free and

  3. #3 peak

    index. According United States Agency for International Development Ethiopia is one of the poorest and most food-insecure countries in the world. And the African capital Addis Ababa is

  4. #4 peak

    take it more seriously. It goes without saying that being in the middle of a pandemic is stressful to everyone, however, people with chronic diseases such as hypo- or hyperthyroidism, Grave’s diseases, goiter, thyroid cancer,

  5. #5 peak

    :⏎- Vitamin E (α-tocopherol)⏎- Causes and Risks for Vitamin E Deficiency⏎- Symptoms of Vitamin E Deficiency⏎- Food Sources of Vitamin E⏎- Vitamin K⏎- Ris

  6. #6 peak

    1, the term Kabinett was officially noted in German wine law, and it was given its current definition which applies to wines which are light and non-chaptalized – chaptalization is the process of adding sugar to unfer

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: how to share a shared identity between people who are super super super super super super super super super super super super super super super super super super super super super super super super super super super super super super super super super super super super super super

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

    team studied the copper chaperone protein Atx1, which provides a good model of copper metabolism in animal cells. "We wondered what the drug tetrathiomolybdate did to copper chaperones -- proteins charged with safely ferry

  2. #2 peak

    ding Light On The Workings Of An Anti-Cancer Drug⏎Crystal structure shows how drug disables harmful copper in cells⏎The copper sequestering drug tetrathiomolybdate (TM) has been shown in studies to be effective in

  3. #3 peak

    In 1993, the foundations to the Dhamma Talaka Peace Pagoda were laid in Birmingham, England. Supported by the Birmingham Buddhist Vihara, the pagoda underwent a large amount of planning. The circumference of the

  4. #4 peak

    experiment involved 42 severely depressed patients who had already agreed to undergo electroconvulsive therapy (ECT), which is typically used as a last resort when all other measures have failed.

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. L19:826 excites
  2. L23:391 suppresses
  3. L18:1369 excites
  4. L20:126 suppresses
  5. L17:2025 excites
  6. L19:2954 suppresses
  7. L15:2791 excites
  8. L16:295 excites
  9. L15:2058 suppresses
  10. L18:1286 excites
  11. L19:2114 suppresses
  12. L18:1381 excites
  13. L16:1560 excites
  14. L22:289 excites
  15. L15:407 suppresses
  16. L22:693 suppresses
  17. L20:1765 suppresses
  18. L15:2411 suppresses
  19. L19:1637 suppresses
  20. L19:2607 suppresses
  21. L19:3025 suppresses
  22. L22:1678 suppresses
  23. L16:2907 excites
  24. L18:3033 excites

Feeds later layers

  1. L26:406 strongly excites
  2. L27:714 strongly excites
  3. L27:2800 excites
  4. L27:860 excites
  5. L27:918 excites
  6. L27:873 excites
  7. L26:521 weakly excites
  8. L26:781 weakly excites
  9. L26:786 weakly excites
  10. L26:1098 weakly excites
  11. L27:2042 weakly excites
  12. L26:536 weakly suppresses
  13. L27:926 weakly excites
  14. L27:685 weakly suppresses
  15. L26:215 weakly excites
  16. L26:2103 weakly excites
  17. L26:1233 weakly excites
  18. L26:754 weakly suppresses
  19. L27:2409 weakly excites
  20. L27:219 weakly excites
  21. L26:1868 weakly excites
  22. L27:1290 weakly suppresses
  23. L27:1592 weakly excites
  24. L26:983 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 -11.67 and -10.781 values between -8.09 and -7.191 values between -6.30 and -5.401 values between -5.40 and -4.501 values between -3.61 and -2.7110 values between -2.71 and -1.8230 values between -1.82 and -0.921691 values between -0.92 and -0.022100 values between -0.02 and 0.8737 values between 0.87 and 1.7714 values between 1.77 and 2.678 values between 2.67 and 3.563 values between 3.56 and 4.462 values between 4.46 and 5.353 values between 5.35 and 6.252 values between 6.25 and 7.151 values between 7.15 and 8.041 values between 8.94 and 9.831 values between 23.28 and 24.17offnegativepositiveoften

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. 不
  2. not
  3. doesn
  4. don
  5. doesn
  6. 不
  7. not
  8. didn
  9. не
  10. 都不

Suppresses

  1. Gratuit
  2. tré
  3. meer
  4. rawn
  5. ʀ
  6. devil
  7. anova
  8. -pencil
  9. JV
  10. asename