Neuron Explorer

Qwen3-0.6B

'side' in 'side effects'

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 'side' when it is followed by 'effects', as in 'side effects of medication'. When turned up, the model kept producing the word 'effect' again and again.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    cause twin problems, as the contamination needs to be removed before water can be piped into homes and the nitrates which fertilise agricultural crops have the unwanted side effect of accelerating the growth

  2. #2 peak

    activates it. When she gave morphine to mice that couldn't make beta-arrestin, they were still numb to pain, but a lot of the negative side effects of the drug were missing. They didn't build tolerance to the drug.

  3. #3 peak

    August, 2009⏎A number of hospitals in London are using yoga therapy as an alternative to antidepressant drugs for patients who are wary of the side effects of medication.⏎Mental health charities are seeing a huge increase in people seeking

  4. #4 peak

    is conveyed in the meme relating to internet privacy. The law that was reverted prevented internet providers from selling<|endoftext|>Chemotherapy is known to come with a long list of side effects -- from debilitating nausea and hair loss to extreme fatigue -- and in many cases

  5. #5 peak

    (disease, anxiety, etc.)⏎- Have a normalizing effect, improving the function of many body systems⏎- Be nontoxic, causing no significant side effects⏎Since that time, the term adaptogen has been generalized to include herbal

  6. #6 peak

    , doctors need to know if you are taking these products so they can accurately judge whether particular symptoms are related to a medical illness or a side effect from an herbal medicine. Those who take herbal products should also keep a reliable

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 do we at times make something that is not something that is actually something? What does that mean? What do we make here, as a group, in which we all belong, and we all make something, which is something that is

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

    invitation of President Franklin Roosevelt and spent the Second World War years in America. Otto von Habsburg returned to Europe after the Nazi defeat, married Princess Regina of Sachsen-Meiningen, who was working as<|endoftext|>Posted by Signature Medical Group

  2. #2 peak

    :⏎- Blood sugar control⏎- Reduce cholesterol and triglyceride levels⏎- Lower blood pressure⏎- Improve memory and cognition⏎What about cinnamon and HIV⏎Acquired immunodeficiency syndrome (AIDS) is caused by HIV and HIV-

  3. #3 peak

    azzucato. Copyright © 2018. Available from PublicAffairs, an imprint of Perseus Books, LLC, a subsidiary of Hachette Book Group, Inc.⏎The January 2010 edition of The

  4. #4 peak

    uary (latin mortalis)|⏎|raaj (king, rule)||rich, regal, regulate|⏎|sharkara||saccharin, sucrose, sugar|⏎|vak, vaacha||voice, vocal|⏎|ved

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. L22:2377 strongly excites
  2. L22:3036 excites
  3. L21:2697 excites
  4. L18:2809 suppresses
  5. L22:2387 suppresses
  6. L21:2760 excites
  7. L21:2653 excites
  8. L19:1832 suppresses
  9. L22:1955 suppresses
  10. L20:2550 excites
  11. L8:2682 excites
  12. L19:1571 weakly suppresses
  13. L4:1536 weakly suppresses
  14. L15:451 weakly suppresses
  15. L12:2526 weakly excites
  16. L9:473 weakly excites
  17. L6:876 weakly excites
  18. L21:2933 weakly excites
  19. L22:1463 weakly suppresses
  20. L9:2620 weakly suppresses
  21. L16:1888 weakly suppresses
  22. L8:968 weakly excites
  23. L19:1518 weakly suppresses
  24. L19:2690 weakly suppresses

Feeds later layers

  1. L27:2527 excites
  2. L24:1062 excites
  3. L24:3070 excites
  4. L25:312 excites
  5. L26:2777 excites
  6. L25:1424 excites
  7. L27:2969 excites
  8. L25:489 excites
  9. L26:1868 weakly suppresses
  10. L27:1544 weakly excites
  11. L25:2861 weakly excites
  12. L24:2607 weakly suppresses
  13. L27:125 weakly excites
  14. L27:1599 weakly excites
  15. L27:672 weakly excites
  16. L24:1107 weakly excites
  17. L26:2338 weakly excites
  18. L25:1114 weakly excites
  19. L25:1967 weakly excites
  20. L27:1704 weakly excites
  21. L25:1082 weakly excites
  22. L27:127 weakly suppresses
  23. L26:507 weakly excites
  24. L26:1682 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 -11.31 and -10.631 values between -9.95 and -9.271 values between -8.60 and -7.923 values between -7.92 and -7.241 values between -7.24 and -6.561 values between -6.56 and -5.885 values between -5.88 and -5.203 values between -5.20 and -4.524 values between -4.52 and -3.845 values between -3.84 and -3.168 values between -3.16 and -2.4811 values between -2.48 and -1.8026 values between -1.80 and -1.12270 values between -1.12 and -0.442605 values between -0.44 and 0.24855 values between 0.24 and 0.9259 values between 0.92 and 1.6014 values between 1.60 and 2.285 values between 2.28 and 2.965 values between 2.96 and 3.644 values between 3.64 and 4.324 values between 4.32 and 5.003 values between 5.00 and 5.683 values between 5.68 and 6.353 values between 6.35 and 7.031 values between 7.03 and 7.711 values between 7.71 and 8.392 values between 8.39 and 9.071 values between 9.75 and 10.431 values between 11.11 and 11.791 values between 13.83 and 14.511 values between 15.19 and 15.87offnegativepositiveoften

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. effect
  2. effect
  3. Effect
  4. effects
  5. -effect
  6. Effect
  7. _effect
  8. (effect
  9. 效应
  10. effects

Suppresses

  1. ogr
  2. rox
  3. posto
  4. диагност
  5. viện
  6. ICC
  7. 樽
  8. rowned
  9. aptor
  10. asctime