Neuron Explorer

Qwen3-0.6B

'has' or 'have' after a subject

A neuron in layer 25 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 responds to the helper words 'has' and 'have' that come after a subject, as in 'there has been' or 'a collision has happened'. Its negative side reacts to the letters 'Is' inside names like Islay. Steering did not give a clear theme, though turning it down made the model repeat 'would' and 'will'.

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

    of the built environment in addressing climate change is clear as buildings account for almost 40% of global carbon emissions according to the World Green Building Council. There has never been a greater commitment

  2. #2 peak

    see section below for printable documents, Ballotechnics books and related discussion.⏎Suggested Pdf Resources⏎Suggested News Resources⏎Suggested Web Resources⏎Great care has been taken to prepare the information on this page. Elements of the content come from

  3. #3 peak

    can "hash" to the same position. If this occurs, then a collision has happened. How do you handle the collision? Hashing also leads to lots of

  4. #4 peak

    Cathedral and Metropolitan Church of St. Peter’, is not just a hugely significant place of worship, but a gigantic repository of architectural and artistic splendour. There has been a church on this site since around 627 AD, when a

  5. #5 peak

    (TESS) Program⏎There is a gap between scientific information generated by researchers and the needs of society to address today’s increasingly complex environmental problems. And there has been frustration on both sides: researchers want to see the result of their efforts being

  6. #6 peak

    dangerous gases owing to the following properties:⏎Tips for Safe Storage of Compressed Acetylene Gas Cylinders⏎Owing to the above discussed hazards, it has become important to take special care while storing and handling compressed gas cylinders of acety

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 would would will would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would would

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

    436) and published books on rhetoric and logic that established his reputation as a teacher and scholar. The work on logic, Isagoge dialéctica, was widely used as a textbook in northern Europe during

  2. #2 peak

    rochs from the northern shore. They landed first, or rather were carried by wind and stream, upon the little isle of Oronsay, close by Islay; and here for a moment they thought their future abode was to be

  3. #3 peak

    (most importantly Salzach, Mur, Isel and Möll).⏎The national park extends over three Austrian provinces: Salzburg

  4. #4 peak

    together with the Islands and Isletts within tenn leagues thereof" into a county that they named Albem

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. L23:30 suppresses
  2. L21:262 excites
  3. L21:1521 excites
  4. L21:1621 suppresses
  5. L14:149 suppresses
  6. L20:586 suppresses
  7. L19:545 excites
  8. L15:1160 suppresses
  9. L17:227 suppresses
  10. L21:1555 excites
  11. L14:623 excites
  12. L16:466 weakly excites 'is' after a long subject phrase
  13. L17:1627 weakly excites
  14. L22:363 weakly excites
  15. L18:93 weakly suppresses
  16. L16:968 weakly suppresses
  17. L6:918 weakly excites
  18. L15:599 weakly excites
  19. L10:586 weakly excites 'important' and degree words after 'how'
  20. L16:316 weakly suppresses
  21. L16:2536 weakly excites
  22. L20:1618 weakly excites 'be' after 'to' or 'need to'
  23. L17:451 weakly excites
  24. L4:20 weakly suppresses

Feeds later layers

  1. L26:53 weakly excites
  2. L27:604 weakly suppresses
  3. L26:1257 weakly suppresses
  4. L27:938 weakly excites
  5. L26:765 weakly suppresses
  6. L27:565 weakly suppresses
  7. L27:774 weakly suppresses
  8. L26:1587 weakly suppresses
  9. L27:500 weakly suppresses
  10. L26:859 weakly suppresses
  11. L27:650 weakly suppresses
  12. L26:467 weakly suppresses
  13. L26:74 weakly suppresses
  14. L26:1534 weakly suppresses
  15. L27:1174 weakly suppresses
  16. L27:717 weakly suppresses
  17. L27:224 weakly suppresses
  18. L26:1615 weakly suppresses
  19. L26:400 weakly suppresses
  20. L26:304 weakly suppresses
  21. L27:2091 weakly suppresses
  22. L26:1452 weakly excites
  23. L26:781 weakly suppresses
  24. L26:500 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 -17.58 and -16.221 values between -14.86 and -13.502 values between -13.50 and -12.146 values between -12.14 and -10.783 values between -10.78 and -9.426 values between -9.42 and -8.0614 values between -8.06 and -6.7017 values between -6.70 and -5.3426 values between -5.34 and -3.9853 values between -3.98 and -2.62110 values between -2.62 and -1.261480 values between -1.26 and 0.10878 values between 0.10 and 1.46300 values between 1.46 and 2.82249 values between 2.82 and 4.18191 values between 4.18 and 5.54125 values between 5.54 and 6.9099 values between 6.90 and 8.2683 values between 8.26 and 9.6250 values between 9.62 and 10.9845 values between 10.98 and 12.3429 values between 12.34 and 13.7025 values between 13.70 and 15.0619 values between 15.06 and 16.4219 values between 16.42 and 17.7821 values between 17.78 and 19.1410 values between 19.14 and 20.5015 values between 20.50 and 21.867 values between 21.86 and 23.226 values between 23.22 and 24.585 values between 24.58 and 25.935 values between 25.93 and 27.292 values between 27.29 and 28.652 values between 28.65 and 30.012 values between 32.73 and 34.091 values between 34.09 and 35.451 values between 35.45 and 36.81offnegativepositiveoften

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. 有一次
  2. ちょうど
  3. rometer
  4. gone
  5. onen
  6. уча
  7. 要有
  8. 嫁给
  9. naissance
  10. ffe

Suppresses

  1. 們
  2. 的人都
  3. amiento
  4. can
  5. 您可以
  6. 我可以
  7. leri
  8. would
  9. nào
  10. will