Neuron Explorer

Qwen3-0.6B

medical advice and health disclaimers

A neuron in layer 14 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 warnings, like 'seek the advice of your doctor', 'healthcare provider', or 'sleep schedule'. It is most active in medical boilerplate. Turning it up or down did not clearly change the text toward health topics.

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

    900 – 14 August 1958) was a French physicist, who shared the 1935 Nobel Prize in Chemistry with his wife Irène Jol

  2. #2 peak

    they can be dangerously habit forming and result in grogginess the following day. Here are some safe techniques to promote good sleep hygiene:⏎- Establish and maintain a consistent sleep schedule.<|endoftext|>A new tool to identify the calls of bat species could help

  3. #3 peak

    to Athens, the metropolis of Greece.⏎- an indigenous or resident of Athens.⏎Old English Atheniense (plural noun), from Latin Atheniensis, from

  4. #4 peak

    If you think you may have a medical emergency, call your doctor or (in the United States) 911 immediately. Always seek the advice of your doctor before starting or changing treatment. Medical professionals who provide responses to health-related

  5. #5 peak

    professional medical advice. CALL YOUR HEALTHCARE PROVIDER IMMEDIATELY IF YOU THINK YOU MAY HAVE A MEDICAL EMERGENCY. Always seek the advice of your physician or other qualified health provider prior to starting any new treatment or with any

  6. #6 peak

    Talk with your healthcare provider for assistance or check out your local health department<|endoftext|>Indigenous Websites:Information Networks

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: the rose and the rose's head and the rose's head and the door and the mirror and the teahouse and the teahouse and a lot of people and a lot of people and a lot of a lot of a lot of

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

    64, the 24th Amendment let vote those who could not pay a poll tax.⏎In 1965, the Voting Rights Act removed literacy tests.⏎On JUNE 22, 1970, President

  2. #2 peak

    synthetic aperture radar, which can provide detailed photographs even through cloud cover. Although still in development, the Global Hawk was used during operations over Afghanistan

  3. #3 peak

    who died trying to climb the wall to escape the Iron Curtain. At noon, Berlin observed a minute of silence in memory of the victims. Buses and trains stopped for the tribute. The Berlin Wall was erected in 1961 on

  4. #4 peak

    prove your identity was literally found at your fingertips. Fingerprints and signatures used to be a trustworthy component of your acceptance or approval – especially in

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. L13:1265 excites
  2. L12:421 suppresses
  3. L12:2442 suppresses
  4. L5:1338 excites
  5. L4:1185 weakly excites
  6. L12:2027 weakly excites
  7. L12:681 weakly excites
  8. L0:1525 weakly suppresses
  9. L8:1196 weakly excites
  10. L13:2124 weakly suppresses
  11. L13:1852 weakly excites
  12. L8:1804 weakly excites
  13. L11:853 weakly suppresses
  14. L12:2500 weakly suppresses
  15. L11:0 weakly suppresses
  16. L5:972 weakly suppresses
  17. L2:539 weakly suppresses
  18. L12:1469 weakly suppresses
  19. L13:1891 weakly excites
  20. L13:1520 weakly suppresses
  21. L5:714 weakly excites
  22. L13:305 weakly excites
  23. L8:2753 weakly suppresses
  24. L13:2733 weakly suppresses

Feeds later layers

  1. L15:939 suppresses
  2. L20:2065 suppresses
  3. L16:2407 suppresses
  4. L17:1717 suppresses
  5. L17:332 suppresses
  6. L16:1343 suppresses
  7. L16:1129 weakly suppresses
  8. L16:1448 weakly suppresses
  9. L16:1340 weakly suppresses
  10. L16:275 weakly suppresses
  11. L15:1831 weakly suppresses
  12. L15:420 weakly suppresses
  13. L20:573 weakly suppresses
  14. L17:2656 weakly suppresses
  15. L15:1529 weakly excites
  16. L17:57 weakly suppresses
  17. L16:498 weakly excites
  18. L19:802 weakly excites unclear: word before a company or brand name
  19. L25:78 weakly excites
  20. L15:1351 weakly suppresses
  21. L15:30 weakly excites
  22. L16:915 weakly excites
  23. L15:481 weakly excites
  24. L20:2162 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 -1.46 and -1.361 values between -1.36 and -1.271 values between -1.27 and -1.172 values between -1.17 and -1.074 values between -0.97 and -0.886 values between -0.88 and -0.783 values between -0.78 and -0.6815 values between -0.68 and -0.5915 values between -0.59 and -0.4944 values between -0.49 and -0.3961 values between -0.39 and -0.30105 values between -0.30 and -0.20312 values between -0.20 and -0.101104 values between -0.10 and -0.001509 values between -0.00 and 0.09329 values between 0.09 and 0.19136 values between 0.19 and 0.2972 values between 0.29 and 0.3958 values between 0.39 and 0.4836 values between 0.48 and 0.5826 values between 0.58 and 0.6815 values between 0.68 and 0.7712 values between 0.77 and 0.878 values between 0.87 and 0.975 values between 0.97 and 1.078 values between 1.07 and 1.164 values between 1.16 and 1.261 values between 1.26 and 1.364 values between 1.36 and 1.453 values between 1.45 and 1.552 values between 1.55 and 1.651 values between 1.65 and 1.742 values between 1.74 and 1.841 values between 2.04 and 2.132 values between 2.33 and 2.42offnegativepositiveoften

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. że
  2. 俾
  3. AttributeName
  4. 怠
  5. คอย
  6. żu
  7. kest
  8. istra
  9. springfox
  10. ObjectId

Suppresses

  1. 弟
  2. inion
  3. 粉
  4. ريب
  5. khẩu
  6. iece
  7. ills
  8. 大街
  9. 霸
  10. plet