Neuron Explorer

Qwen3-0.6B

words right before 'have' or 'has'

A neuron in layer 21 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 fires on small words like 'not', 'also', 'always', 'may' and 'might' that sit right before a verb such as 'have' or 'has'. It is positive when a sentence is about to say what something has or can do. Steering results were only loosely related, with some 'have' and 'has' repetition.

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

    .⏎Rh (+) positive blood means that your red blood cells have D antigens on their surface. Rh (-) negative blood means that your red blood cells do not have D antigens on their surface.⏎Rh (+) blood is much more common

  2. #2 peak

    power solar panels. Sunpower panels are made up of photovoltaic cells that convert sunlight into electrical energy. While Sunpower solar panels have many advantages, they also have a few disadvantages.⏎Sunpower solar panels are a premium product and as such

  3. #3 peak

    . For in-depth answers, consider the following resources.⏎Stop By the Tutoring Center⏎Most high schools will not have a tutoring center, but colleges will almost always have an on-site tutoring center. At a university, the tutoring center is offered

  4. #4 peak

    ), there is usually more personal information included in the newspaper than in official documents.⏎Almost every city has a newspaper, and many have more than one. Large cities may have multiple daily papers, while rural areas might have a weekly paper covering

  5. #5 peak

    , brake control function, motor phase number and working mode selection function, protection function (such as current limit, under voltage, overheating protection, etc.), some also integrated drive circuit.⏎Suitable for a certain speed but not high speed accuracy requirements

  6. #6 peak

    to conduct electricity on the other side. A single layer PCB is useful for simpler electronic devices and is a cheaper option.⏎- Double Layer PCB – A double layer PCB has a copper layer on both sides of the board. This type of PCB has

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: it's the first real product I have ever been the first product to there of there there there there, so there there there there, there there there there, there there there there, there there there there, there there there there, there

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

    avoid this problem, follow the “10% rule”: increase mileage no more than a 10% per week. Keep in mind, though, that everyone is different and that even 10% may be too much of a weekly

  2. #2 peak

    are an academic or someone in academic publishing, have you ever wondered how many academic journals are there in the world? Well, I do.⏎It is estimated that there are currently more than 30,000 academic journals, and the

  3. #3 peak

    counterparts. It's common to experience extreme sadness, anger, guilt and anxiety about future pregnancies. There is no "typical" timeframe for emotional recovery; every woman experiences the grieving process in her own way and travels the road to healing at her

  4. #4 peak

    increase their happiness, it also increases our own. Studies amongst different cultures and ages have consistently found that individuals who spend money on others report higher levels of happiness than those who spend money on themselves. Researchers suggest that the reward we experience from helping others

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. L20:1503 strongly excites
  2. L20:1404 excites
  3. L19:2798 excites
  4. L20:1727 excites
  5. L19:504 suppresses
  6. L20:1932 excites
  7. L20:2696 suppresses
  8. L20:1596 suppresses
  9. L17:2314 suppresses
  10. L18:2846 suppresses
  11. L5:2275 suppresses
  12. L20:578 excites
  13. L3:520 suppresses
  14. L20:262 weakly excites
  15. L16:2010 weakly suppresses
  16. L17:2820 weakly excites
  17. L17:2763 weakly excites
  18. L18:991 weakly suppresses
  19. L20:2389 weakly suppresses
  20. L18:2119 weakly suppresses
  21. L18:1822 weakly suppresses
  22. L20:554 weakly suppresses
  23. L1:2538 weakly suppresses
  24. L12:2121 weakly excites

Feeds later layers

  1. L23:1235 excites
  2. L25:1994 suppresses
  3. L24:792 excites
  4. L22:238 excites
  5. L23:984 weakly suppresses
  6. L25:613 weakly suppresses
  7. L25:210 weakly excites
  8. L22:1349 weakly suppresses
  9. L24:670 weakly excites
  10. L25:929 weakly excites
  11. L27:756 weakly excites
  12. L22:1381 weakly suppresses
  13. L23:1500 weakly suppresses
  14. L23:2263 weakly suppresses
  15. L24:319 weakly suppresses
  16. L22:1671 weakly suppresses
  17. L23:628 weakly suppresses
  18. L24:1342 weakly suppresses
  19. L24:1263 weakly excites
  20. L24:1580 weakly suppresses
  21. L27:409 weakly excites
  22. L23:815 weakly excites
  23. L23:1090 weakly suppresses
  24. L24:2253 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 -7.24 and -6.801 values between -5.90 and -5.461 values between -5.46 and -5.012 values between -5.01 and -4.562 values between -4.56 and -4.124 values between -4.12 and -3.675 values between -3.67 and -3.233 values between -3.23 and -2.7813 values between -2.78 and -2.3320 values between -2.33 and -1.8930 values between -1.89 and -1.4465 values between -1.44 and -0.99158 values between -0.99 and -0.55956 values between -0.55 and -0.102112 values between -0.10 and 0.35368 values between 0.35 and 0.7962 values between 0.79 and 1.2426 values between 1.24 and 1.6821 values between 1.68 and 2.1317 values between 2.13 and 2.588 values between 2.58 and 3.028 values between 3.02 and 3.474 values between 3.47 and 3.926 values between 3.92 and 4.364 values between 4.36 and 4.812 values between 4.81 and 5.254 values between 5.25 and 5.701 values between 6.15 and 6.591 values between 7.04 and 7.491 values between 7.49 and 7.931 values between 9.72 and 10.161 values between 10.16 and 10.61offnegativepositiveoften

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. had
  2. has
  3. 具有
  4. have
  5. Owned
  6. having
  7. имеет
  8. .has
  9. 拥有
  10. يط

Suppresses

  1. ⼦
  2. 鼷
  3. ⽬
  4. there
  5. 间的
  6. There
  7. there
  8. cps
  9. THERE
  10. 岁的