Neuron Explorer

Qwen3-0.6B

'in' after a noun phrase

A neuron in layer 23 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 'in' when it follows a noun and introduces a place or setting, such as 'water in a bottle' or 'solar systems in the United States'. Turning it down made the model keep repeating the word 'water'.

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

    chucked in the bin and head straight for our landfill.⏎There are lots of products you can switch to:⏎- Instead of bottled water, change for tap water in a water bottle⏎- Swap your hair washing products for shampoo and conditioner bars⏎

  2. #2 peak

    of the global average, and nearly 700 million of its citizens live in regions considered highly water-stressed. Meanwhile, groundwater depletion has been so significant in the areas around Beijing that parts of the city are falling into the earth by more

  3. #3 peak

    Rep⏎Authors: Altman EA, Lee KL, Hecht CA, Hampton KE, Moreno G, Patel AI⏎Examining Policies Providing Access to Free Water in Schools and the Health Impact of an Alternative to Sugar-Sweetened BeveragesThis

  4. #4 peak

    is clean, with water or carbon dioxide being the only by-product.⏎While high capital costs have been an impediment to the use of fuel cells to power homes in the United States, several businesses use stationary fuel cell systems at operational facilities. As

  5. #5 peak

    can reach such aims. UN-Habitat works to increase global energy access with special emphasis on women and girls and to spread renewable energy technologies to enhance energy supply in urban areas. We particularly support conceptual and analytical awareness and understanding on gender, poverty

  6. #6 peak

    perceptible spurt in crop productivity, which has, of late, tended to stagnate at many places.⏎The saving on water, which is rapidly turning scarce in most paddy-growing tracts, can be 30 to 40

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: water is water, water is water, water is water. In the water water water water water water water water water water water water... water water water water... water water water water... water water water water... water water water water... water

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

    options contract has the following properties or attributes:⏎- Expiration Date⏎- Symbol or Underlying⏎- Strike Price⏎- Type: Call or Put⏎The contract has

  2. #2 peak

    each leg measures about 23 inches. Males are typically slightly larger than females. Both sexes have gills that give the crabs the ability to breathe under water via gas exchange facilitated by the under-body set of small legs.⏎Breeding

  3. #3 peak

    some rights in exchange. This payment is called the premium. There are a few other important properties of this contract that will be explained here: Expiration Date, Underlying, Strike Price, option type: PUT or CALL⏎An

  4. #4 peak

    make that centuries now.⏎It’s true that salt water is a better conductor than fresh water but both environments need anode protection for boats that have any metal parts under water. Fresh water has sufficient impurities that it too conducts electricity, and therefore

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. L21:1935 suppresses
  2. L16:1807 suppresses 'in' after a motion verb
  3. L8:351 suppresses 'in' after 'high' or 'dressed'
  4. L4:289 excites
  5. L20:1574 excites
  6. L5:490 excites
  7. L17:706 excites
  8. L16:2259 weakly excites
  9. L10:525 weakly suppresses
  10. L17:3014 weakly suppresses
  11. L4:2603 weakly suppresses
  12. L9:1025 weakly excites
  13. L21:2644 weakly suppresses
  14. L10:132 weakly excites
  15. L14:1625 weakly excites the word 'in' after a noun
  16. L18:483 weakly excites
  17. L16:1557 weakly excites 'in' meaning 'in this case' or 'in between'
  18. L5:729 weakly suppresses
  19. L16:1913 weakly excites
  20. L19:1354 weakly excites
  21. L3:862 weakly suppresses
  22. L18:3057 weakly suppresses
  23. L4:1966 weakly suppresses
  24. L4:1240 weakly excites

Feeds later layers

  1. L27:2466 weakly suppresses
  2. L25:2356 weakly suppresses
  3. L24:1366 weakly excites
  4. L24:659 weakly suppresses
  5. L24:1989 weakly excites
  6. L24:708 weakly suppresses
  7. L25:3030 weakly suppresses 'Sub' starting a word or name
  8. L25:780 weakly excites opening parenthesis before an acronym
  9. L26:515 weakly suppresses
  10. L24:2374 weakly suppresses
  11. L27:1023 weakly excites
  12. L26:2281 weakly excites
  13. L26:2255 weakly suppresses
  14. L24:1457 weakly excites
  15. L26:2006 weakly suppresses
  16. L24:2961 weakly suppresses
  17. L26:2038 weakly suppresses
  18. L24:239 weakly excites
  19. L27:1922 weakly suppresses
  20. L27:630 weakly suppresses
  21. L27:1916 weakly suppresses capital P starting a word
  22. L24:3043 weakly suppresses
  23. L24:1365 weakly suppresses
  24. L26:1800 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 -28.39 and -26.492 values between -16.96 and -15.061 values between -15.06 and -13.153 values between -11.25 and -9.341 values between -9.34 and -7.4410 values between -7.44 and -5.5317 values between -5.53 and -3.6335 values between -3.63 and -1.722841 values between -1.72 and 0.19817 values between 0.19 and 2.0977 values between 2.09 and 4.0037 values between 4.00 and 5.9018 values between 5.90 and 7.8116 values between 7.81 and 9.717 values between 9.71 and 11.629 values between 11.62 and 13.522 values between 13.52 and 15.432 values between 15.43 and 17.332 values between 17.33 and 19.246 values between 19.24 and 21.141 values between 21.14 and 23.052 values between 23.05 and 24.951 values between 45.91 and 47.81offnegativepositiveoften

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. bas
  2. @⏎
  3. .Sleep
  4. 山村
  5. 又能
  6. ');");⏎
  7. asket
  8. remote
  9. ↾
  10. (provider

Suppresses

  1. water
  2. Water
  3. water
  4. 水
  5. воды
  6. água
  7. Water
  8. _water
  9. WATER
  10. agua