Neuron Explorer

Qwen3-0.6B

'either' before an 'or' choice

A neuron in layer 7 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

The neuron fires on the word 'either' (and 'else' after 'or') when the sentence sets up a choice between two options. When turned up, the model wrote an 'either it's raining, or it's snowing' style line, so the link is visible there. Its strongest negative cases are the word 'and' linking a job title to a name.

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

    to get rid of them. There are many different kinds of dog waste, but most of them are not biodegradable.⏎So what do you do?⏎You either throw away the poop or try to compost it.⏎Composting Dog Poop Is

  2. #2 peak

    of a monster adorned the magazine’s cover. The editorial opined: “The rich world has a clear choice: learn from the mistakes of the past, or else watch Leviathan grow into a true monster.” In a more recent issue, dedicated

  3. #3 peak

    correct voltage is delivered to the device.⏎But what if you already ran your wires and you can’t run a second set to sense the remote voltage? You either design in a regulator on the device side (a good idea regardless) and source

  4. #4 peak

    effective solutions to ensure that we have educated adults entering the work force, without the funds available many will not get the education and the country will suffer. We are either part of the problem or<|endoftext|>People have explored the infinite abyss past Earth’s ambiance

  5. #5 peak

    many passionate musicians who lived in his building and made him think about how the importance of the sound of music is.⏎When he said “In those days it was either live with music or die with noise” means as a person we need to choose

  6. #6 peak

    claims that gift giving creates something called a “debt-balance”, so to prevent negative feelings, gifts must be repaid creating a cycle of gift giving.or else you’ll be interpreted as someone who doesn’t appreciate the relationship with others.⏎-

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: "I'm a bit of a writer, and I'm not sure if I'll make it out. I'm writing this article, and I'm looking to see if we'll be together. I'm writing the article, and I'm

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

    ift Catcher selected as 2008 Tech Awards Laureate by the Tech Museum of Innovation⏎PAN's air monitoring device, invented by chemist and senior scientist Dr. Susan Kegley, was designated a 200

  2. #2 peak

    1⏎NARRATOR: Lying just beneath everyday reality is a breathtaking world, where much of what we perceive about the universe is wrong. Physicist and best-selling author Brian Greene takes you on a journey that bends the rules of human

  3. #3 peak

    losses|⏎The Siege of Jerusalem in the year 70 AD was the decisive event of the First Jewish-Roman War. The Roman army, led by the future Emperor Titus, with Tiberius Julius Alexander as his second-in-command,

  4. #4 peak

    summer (and winter) temperature and unreliable vehicles that allowed the integrity of the clinical environment to be compromised by letting in dust.⏎Froede, along with teammate and mechanical engineering doctorate student Bryan Lewis, is a member of the Global Engineering Team

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. L6:1639 excites
  2. L5:2609 suppresses
  3. L6:2985 weakly suppresses
  4. L6:2499 weakly suppresses
  5. L6:1773 weakly excites
  6. L6:2079 weakly suppresses
  7. L5:1097 weakly suppresses
  8. L2:1315 weakly excites 'of' after nouns like 'variation' and 'section'
  9. L4:88 weakly suppresses
  10. L1:3004 weakly suppresses
  11. L1:1738 weakly excites
  12. L0:2752 weakly suppresses
  13. L6:2183 weakly suppresses
  14. L3:1407 weakly excites
  15. L5:664 weakly excites
  16. L6:1675 weakly excites
  17. L1:2430 weakly excites
  18. L3:2229 weakly excites
  19. L6:3033 weakly suppresses
  20. L6:2216 weakly excites
  21. L6:662 weakly suppresses
  22. L5:2964 weakly excites
  23. L1:2940 weakly suppresses
  24. L0:1772 weakly suppresses

Feeds later layers

  1. L8:2786 suppresses
  2. L10:2097 suppresses
  3. L9:1065 suppresses
  4. L10:1481 suppresses
  5. L8:386 suppresses
  6. L9:434 suppresses comma after an adjective in a list
  7. L8:1049 weakly suppresses
  8. L14:1337 weakly suppresses
  9. L8:489 weakly suppresses
  10. L8:303 weakly suppresses
  11. L11:414 weakly suppresses
  12. L8:2002 weakly excites
  13. L9:493 weakly suppresses
  14. L8:2921 weakly excites
  15. L12:1670 weakly suppresses
  16. L11:91 weakly suppresses
  17. L8:1345 weakly suppresses
  18. L13:1270 weakly suppresses
  19. L8:2143 weakly suppresses
  20. L12:2971 weakly suppresses
  21. L23:288 weakly excites
  22. L26:1976 weakly excites
  23. L14:2023 weakly suppresses
  24. L19:1020 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 -3.03 and -2.931 values between -2.28 and -2.171 values between -1.96 and -1.853 values between -1.64 and -1.532 values between -1.31 and -1.211 values between -1.21 and -1.101 values between -1.10 and -0.992 values between -0.99 and -0.881 values between -0.88 and -0.784 values between -0.78 and -0.678 values between -0.67 and -0.5611 values between -0.56 and -0.4529 values between -0.45 and -0.35117 values between -0.35 and -0.24342 values between -0.24 and -0.13833 values between -0.13 and -0.021336 values between -0.02 and 0.08760 values between 0.08 and 0.19287 values between 0.19 and 0.30128 values between 0.30 and 0.4131 values between 0.41 and 0.514 values between 0.51 and 0.622 values between 0.62 and 0.731 values between 0.84 and 0.941 values between 1.05 and 1.161 values between 1.16 and 1.27offnegativepositiveoften

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. 陂
  2. /Gate
  3. /");⏎
  4. ᴄ
  5. けて
  6. 릅
  7. paddle
  8. 蚝
  9. Violation
  10. ";//

Suppresses

  1. 著名的
  2. 友
  3. �回
  4. older
  5. acclaimed
  6. 经典的
  7. friend
  8. recently
  9. then
  10. 闻名