Neuron Explorer

Qwen3-0.6B

weight loss and metabolism claims

A neuron in layer 25 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires in health articles about diet, fat burning, metabolism and calories, often on words like 'for' or 'fat' in phrases such as 'low-carb diet to be better for weight loss'. When it was turned up, the model drifted into unrelated text, so the link to diet only shows in the examples.

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

    5⏎Paperback ISBN: 9781503603844⏎Ebook ISBN: 9781503604513⏎Request Review/

  2. #2 peak

    This will help you keep your metabolism at a high level for the rest of the day. Research has shown that people who eat a good breakfast

  3. #3 peak

    burns fats and reduces weight⏎Coffee is a caffeine-rich drink; caffeine increase fat burning and improves metabolism rate. In addition, it decreases appetite which is helpful to

  4. #4 peak

    has been shown to help increase brown fat in mice and, in a separate study, was found to increase calorie burn and reduce obesity risk in mice.⏎Apples Are Diverse⏎The apple

  5. #5 peak

    change in diet pattern keeps you full for long and promote metabolism, which further lead to weight loss. Several evidences have proven low-carb diet to be better for quick weight loss, when compared to low-fat diet.⏎Although low-carb diet

  6. #6 peak

    make for a good story.<|endoftext|>The rowing machine might not be as popular as the treadmill at your gym, but rowing is a great alternative to running for your workout. Not only does rowing burn calories and strengthen your heart, it

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 to get the same matching again after again again again again again again again again again again again again again again again again again again again again again again again again again again again again again again again again again again again again again again again again

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

    Physicians and researchers who specialize in the study, diagnosis, treatment, and prevention of cancer are called oncologists.⏎What causes cancer?⏎Cancer is ultimately the result of cells that uncontrollably grow and do not die. Normal cells in the body

  2. #2 peak

    limits fall under the federal Clean Air Act because of their potential to cause or contribute to excessive ozone levels which violate the National Ambient Air Quality Standards. Ozone is an environmental pollutant that contributes to smog.⏎Smog caused by VOCs not

  3. #3 peak

    Dissertation Fellowship, has spent the past year conducting research on the role of ozone sources on both homogeneous and heterogeneous chemistry and their effects on the formation of toxic gases and nano-particles in buildings. In layman’s terms, he

  4. #4 peak

    raise concerns about how long it will take for China’s “war on pollution” to pay off with public health benefits in terms of exposure to fine particulates known as PM2.5, mainly from coal.⏎As it stands now, only about

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. L11:1547 weakly excites
  2. L21:733 weakly excites
  3. L4:267 weakly excites
  4. L13:1320 weakly suppresses
  5. L12:573 weakly suppresses
  6. L4:2991 weakly excites
  7. L8:2220 weakly excites
  8. L5:1467 weakly excites
  9. L13:266 weakly suppresses
  10. L9:719 weakly excites
  11. L16:2146 weakly excites
  12. L23:387 weakly suppresses
  13. L22:2221 weakly suppresses
  14. L7:170 weakly excites
  15. L13:1155 weakly excites
  16. L0:1805 weakly suppresses
  17. L6:2080 weakly excites
  18. L18:320 weakly suppresses
  19. L10:1894 weakly suppresses
  20. L9:207 weakly suppresses
  21. L2:692 weakly excites
  22. L17:2355 weakly suppresses
  23. L4:42 weakly suppresses
  24. L16:1557 weakly excites 'in' meaning 'in this case' or 'in between'

Feeds later layers

  1. L27:2926 weakly suppresses
  2. L27:2949 weakly excites
  3. L26:268 weakly excites
  4. L27:369 weakly suppresses
  5. L26:895 weakly suppresses
  6. L26:1427 weakly suppresses
  7. L26:1506 weakly excites
  8. L26:1469 weakly excites
  9. L27:2319 weakly excites
  10. L26:3047 weakly suppresses
  11. L26:2504 weakly excites
  12. L27:1204 weakly suppresses
  13. L27:1466 weakly suppresses
  14. L27:654 weakly suppresses
  15. L27:1985 weakly suppresses
  16. L26:1389 weakly excites
  17. L26:2159 weakly excites
  18. L27:2304 weakly suppresses
  19. L27:35 weakly excites
  20. L26:398 weakly suppresses
  21. L27:629 weakly suppresses
  22. L26:1911 weakly excites
  23. L27:1987 weakly excites
  24. L27:1655 weakly excites

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 -16.08 and -15.231 values between -15.23 and -14.381 values between -14.38 and -13.531 values between -12.68 and -11.831 values between -10.98 and -10.131 values between -9.28 and -8.431 values between -8.43 and -7.571 values between -6.72 and -5.871 values between -5.87 and -5.027 values between -5.02 and -4.1712 values between -4.17 and -3.3223 values between -3.32 and -2.4737 values between -2.47 and -1.62114 values between -1.62 and -0.772098 values between -0.77 and 0.081424 values between 0.08 and 0.9394 values between 0.93 and 1.7833 values between 1.78 and 2.6320 values between 2.63 and 3.4813 values between 3.48 and 4.3310 values between 4.33 and 5.184 values between 5.18 and 6.032 values between 6.03 and 6.881 values between 6.88 and 7.732 values between 8.58 and 9.432 values between 9.43 and 10.281 values between 10.28 and 11.131 values between 11.13 and 11.991 values between 17.09 and 17.94offnegativepositiveoften

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. 藓
  2. 事
  3. AFF
  4. 绦
  5. 生
  6. ech
  7. itures
  8. 菊
  9. ine
  10. bú

Suppresses

  1. (Symbol
  2. CodeAt
  3. SQ
  4. ồi
  5. SQ
  6. ��
  7. омер
  8. фон
  9. grim
  10. izar