Neuron Explorer

Qwen3-0.6B

"any of" before "these" or "the"

A neuron in layer 19 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 "of" right after "any" or "one", as in "any of these symptoms" or "one of the groups". When it was turned up, the model tended to drift into repeating phrases like "one single" or "a single drop".

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

    to anatomy, physiology, pathology, and therapy. This⏎section contains a variety of research hypotheses waiting to be pursued.⏎If you have expertise or interest in any of these areas and would⏎like to participate in our research programs, please contact us

  2. #2 peak

    , this medication could cause more serious side effects, including dizziness, hearing loss, or an allergic reaction. You should seek immediate medical assistance if you notice any of these symptoms.⏎Dosage information relates to the recommended dosage and frequency of drugs a

  3. #3 peak

    any ingredient in Ampicillin or any other penicillin antibiotic (e.g., amoxicillin).⏎Contact your doctor or health care provider right away if any of these apply to you.⏎Some medical conditions may interact with Ampicillin . Tell

  4. #4 peak

    the forces of supply and demand. Any shift in one of these results in a price change that signals a corresponding shift in the other. Then

  5. #5 peak

    see if mortality or disease in any of the groups differed.⏎The individual pieces of data they found from dietary analysis were:

  6. #6 peak

    with the movement of the fluid. If any of these values increase (resp. decrease), the power also increases (resp. decreases

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: one single single one single single single one single single single one single single single single single single single one single one single one single one one single one single one one single one single one one single one single one single one single one one single one

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

    the outcome, I don't think we would have had such a comprehensive service from any other law firm, and you took the worry away...I do not regret a single second of the whole process, apart from the bit before you got involved.⏎

  2. #2 peak

    uncontrolled terrestrial conditions according to evolutionists, still cannot be synthesised even in the most sophisticated, high-tech laboratories of the 20th century. Not a single "transitional form", creatures which are supposed to show the gradual evolution of

  3. #3 peak

    B.C.) ruled over Judea for more than 30 years and built some of the ancient world’s most impressive monuments, it might seem odd that not a single coin has been found that bears the king’s likeness. After all, most

  4. #4 peak

    power plant? Ten minutes ago, I got a newsletter from the unmanned-vehicles industry about all the cool things robots are doing to help Japan. It has not a word about the nuclear reactors. That's disgraceful.⏎To head off the next

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. L16:752 excites
  2. L17:1644 suppresses
  3. L11:1095 weakly suppresses
  4. L14:1285 weakly excites
  5. L18:569 weakly excites
  6. L17:2705 weakly excites 'of' in 'one of the' phrases
  7. L1:509 weakly excites 'to' after 'unable'
  8. L7:2272 weakly suppresses
  9. L5:110 weakly suppresses
  10. L6:2681 weakly excites
  11. L3:2726 weakly excites
  12. L14:104 weakly excites
  13. L16:1071 weakly suppresses "any" in questions and negatives
  14. L13:3010 weakly suppresses
  15. L3:2793 weakly suppresses
  16. L1:1424 weakly suppresses
  17. L18:886 weakly suppresses
  18. L4:151 weakly excites
  19. L18:1110 weakly suppresses
  20. L13:1665 weakly suppresses
  21. L1:951 weakly excites
  22. L9:1236 weakly excites
  23. L3:2813 weakly suppresses
  24. L7:469 weakly suppresses

Feeds later layers

  1. L20:1279 strongly suppresses
  2. L25:2737 suppresses 'a' before 'few' or 'couple'
  3. L22:1033 suppresses
  4. L27:556 suppresses
  5. L23:2083 suppresses 'a' meaning one single
  6. L26:1028 suppresses
  7. L24:979 suppresses
  8. L25:572 suppresses
  9. L23:1947 suppresses
  10. L24:2538 suppresses
  11. L22:1070 suppresses
  12. L25:2580 weakly suppresses
  13. L23:617 weakly suppresses
  14. L20:2853 weakly suppresses
  15. L22:2656 weakly suppresses
  16. L25:549 weakly excites
  17. L21:1120 weakly suppresses
  18. L23:97 weakly suppresses
  19. L25:778 weakly suppresses
  20. L27:789 weakly suppresses
  21. L24:823 weakly suppresses
  22. L27:180 weakly excites
  23. L22:1176 weakly suppresses
  24. L22:2238 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.

2 values between -4.47 and -4.071 values between -4.07 and -3.672 values between -3.67 and -3.271 values between -2.86 and -2.461 values between -2.46 and -2.061 values between -2.06 and -1.661 values between -1.66 and -1.254 values between -1.25 and -0.8517 values between -0.85 and -0.45698 values between -0.45 and -0.052950 values between -0.05 and 0.36173 values between 0.36 and 0.7640 values between 0.76 and 1.166 values between 1.16 and 1.562 values between 1.56 and 1.972 values between 1.97 and 2.371 values between 3.17 and 3.581 values between 3.58 and 3.981 values between 5.19 and 5.591 values between 5.99 and 6.391 values between 6.80 and 7.201 values between 9.21 and 9.611 values between 11.22 and 11.63offnegativepositiveoften

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. orget
  2. 为代表
  3. ulos
  4. >null
  5. ivement
  6. oured
  7. toUpperCase
  8. 很漂亮
  9. 为主体
  10. 通知书

Suppresses

  1. one
  2. one
  3. -one
  4. 一声
  5. 一丝
  6. single
  7. least
  8. ONE
  9. least
  10. _one