Neuron Explorer

Qwen3-0.6B

unclear: verbs in health and risk claims

A neuron in layer 15 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 words in health, safety and risk sentences, such as 'beneficial', 'security' and 'clump'. Its strongest negative cases are words like 'to' and 'run' where a sentence is about to contrast one idea with another. Turning it down made the model repeat the word 'but' over and over.

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

    odds of getting coronary disease, over-nutrition is detrimental to you given it improves fat retailers. A diet that is definitely way too lower in energy is often beneficial simply because it oversees your urge for food helping decrease how many calories you consume food

  2. #2 peak

    , use blue light. Exposing the eyes to light in the evening prevents the flow of melatonin from starting. By wearing glasses that block blue light allows the flow to start (2). The average time for melatonin to flow (if the

  3. #3 peak

    . This stops the bleeding in your body. If you have a cut or a wound, this is a good thing. But, sometimes, platelets will clump together inside a blood vessel that is injured, swollen (inflamed), or

  4. #4 peak

    much higher levels than others.⏎Take a look at our list of the top 15 most contaminated fish on the market today. Avoid eating these or eat them in very small quantities, unless you are an expectant mother, in which case

  5. #5 peak

    is that most of the problems all came from a larger, systemic problem: the owners and operators of steamships had for five decades taken larger and larger risks to save money - risks to which they had methodically blinded themselves

  6. #6 peak

    s do deliver greater mobility and accessibility to information, but can also increase security risks if the appropriate precautions are not taken. However, digital records can actually provide greater security if properly

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: but, but, but, but, but, but, but, but, but, but, but, but, but, but, but, but, but, but, but, but, but, but, but, but,

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

    O blood. Type A blood, which is more common in Central and Eastern Europe, has A antigens on its surface. People who are type A can donate to either type A or type AB people and can receive blood from type O or type

  2. #2 peak

    of selective toxicity: that certain chemicals that would be toxic to some organisms, e.g., infectious bacteria, would be harmless to other organisms, e

  3. #3 peak

    ers work on the ideology of prevention being the best cure rather than negatively incentivizing the drivers. Research has also found that while negative incentives do work over the short run, it cannot account for human behavioural tendencies. Speeding as such being a very

  4. #4 peak

    classwide discussion in classes both large and small, in the sciences as well as in the humanities.⏎Attending a lecture and taking notes works very well for some students, particularly the students who go on to careers in academia. They’re able

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. L13:2192 excites
  2. L14:1277 excites
  3. L14:2603 excites
  4. L12:2622 excites
  5. L11:2778 excites
  6. L12:2573 suppresses
  7. L14:2135 excites
  8. L13:1348 excites
  9. L13:689 excites
  10. L14:2329 suppresses
  11. L14:111 suppresses
  12. L13:1643 suppresses
  13. L14:864 suppresses
  14. L14:2385 suppresses
  15. L14:480 suppresses
  16. L12:1680 suppresses
  17. L13:2225 weakly excites
  18. L14:2480 weakly suppresses
  19. L14:2764 weakly excites
  20. L13:528 weakly excites
  21. L14:779 weakly suppresses 'than' in comparisons
  22. L14:2990 weakly excites
  23. L14:229 weakly excites
  24. L14:2968 weakly excites

Feeds later layers

  1. L21:478 strongly suppresses
  2. L16:2865 strongly suppresses
  3. L19:2590 strongly suppresses
  4. L27:568 strongly suppresses
  5. L25:476 strongly suppresses comma before 'but' in contrasts
  6. L16:2778 strongly suppresses
  7. L18:2614 strongly suppresses
  8. L16:533 excites
  9. L16:2373 suppresses
  10. L20:1194 suppresses
  11. L18:787 suppresses
  12. L19:2294 suppresses
  13. L20:1764 suppresses
  14. L16:788 suppresses
  15. L16:3060 suppresses
  16. L17:2752 suppresses
  17. L26:1232 suppresses
  18. L16:1084 suppresses
  19. L17:2541 suppresses comma before but also in not only constructions
  20. L20:2787 suppresses
  21. L16:2902 suppresses
  22. L25:1607 weakly suppresses
  23. L19:2919 weakly suppresses
  24. L17:1398 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 -1.69 and -1.611 values between -1.61 and -1.522 values between -1.26 and -1.181 values between -1.18 and -1.092 values between -1.09 and -1.011 values between -1.01 and -0.923 values between -0.92 and -0.832 values between -0.83 and -0.752 values between -0.75 and -0.664 values between -0.66 and -0.585 values between -0.58 and -0.4917 values between -0.49 and -0.4141 values between -0.41 and -0.3278 values between -0.32 and -0.23227 values between -0.23 and -0.15549 values between -0.15 and -0.061213 values between -0.06 and 0.02828 values between 0.02 and 0.11464 values between 0.11 and 0.19243 values between 0.19 and 0.28137 values between 0.28 and 0.3744 values between 0.37 and 0.4523 values between 0.45 and 0.5410 values between 0.54 and 0.622 values between 0.62 and 0.713 values between 0.71 and 0.791 values between 0.88 and 0.971 values between 1.22 and 1.312 values between 1.31 and 1.391 values between 1.65 and 1.74offnegativepositiveoften

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. consist
  2. (vertices
  3. äch
  4. ))*(
  5. 一贯
  6. @endforeach
  7. assorted
  8. (action
  9. consisted
  10. direct

Suppresses

  1. 但
  2. but
  3. but
  4. 但是
  5. 但它
  6. 但却
  7. 但他
  8. 但对于
  9. 但在
  10. 但是对于