Neuron Explorer

Qwen3-0.6B

comma after 'Unlike X' contrasts

A neuron in layer 8 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 comma that follows a phrase starting with 'Unlike', as in 'Unlike running, rowing...', where a contrast is about to be made. It goes quiet, and sometimes pushes the other way, on words like 'usual' in phrases such as 'with his usual thoughtfulness'. Turning it up gave unrelated text, so the link to the steering output is weak.

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

    health. Medicine is to the individual as public health is to the population. While doctors treat illnesses, public health professionals implement health promotion and health education.⏎Unlike medicine, there is no instant gratification with public health. It’s like Michelangelo and

  2. #2 peak

    as “reminiscent of the open cellular convection that is often seen over Earth’s oceans”…but with one significant exception. “Unlike on Earth, where such layers are just above the surface,” he said, “this one is

  3. #3 peak

    shooting spree in Aurora, Colorado. Bullets are so easy to come by that it's clear that huge stockpiles exist throughout the country. But unlike guns, bullets

  4. #4 peak

    ? hole⏎Answerwaist coat tail gate way side arm pit fall out put down cast off spring time line man hole⏎See another brain teaser just like this one...⏎Or, just get a random brain teaser⏎If you become a registered user

  5. #5 peak

    inae in your core, and your deltoids, biceps and brachioradialis in your arms.⏎Kinder to Your Joints⏎Unlike running, rowing is both low-impact and non-weight bearing, so it causes less

  6. #6 peak

    protect the public health.”⏎Talc & asbestos⏎Talc and asbestos are naturally occurring minerals that may be found in close proximity in the earth. Unlike talc, however, asbestos is a known carcinogen, to prevent contamination of talc with

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 know that the best way to do this, the "Bingo" for the math competition. I will look at the articles in this context. So, I have a feeling that the best mathematical results in the areas of both: -

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

    East and the Aegean, manifesting as declining populations, destruction, reduced trade, and significant cultural changes.⏎As usual, fingers have been pointed at climate change and shifting allegiances. But scientists have

  2. #2 peak

    stream was sent 240 feet, almost an entire city block.⏎A Most Inviting Repast⏎"In the evening the ladies of Seattle, with their usual thoughtfulness, spread a most inviting repast in the Council chamber which was highly

  3. #3 peak

    assembly thwarted him. Refusing to accept defeat, he modified a CCFL inverter into an arc lighter, and the process is pretty educational.⏎With his usual impeccable handwriting and schematic drawing skills, [GreatScott!] documents that his CCFL

  4. #4 peak

    manner of charging your battery system is important in maximizing battery life and performance.<|endoftext|>A new set of MCQ Questions with Answer in the Computer Networking category. As usual there are 10 Objective

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. L7:530 suppresses
  2. L7:597 excites
  3. L5:1552 weakly suppresses
  4. L7:49 weakly suppresses
  5. L7:1705 weakly excites
  6. L5:737 weakly suppresses
  7. L7:526 weakly suppresses
  8. L7:1970 weakly suppresses
  9. L7:2071 weakly excites
  10. L7:1593 weakly suppresses
  11. L7:1361 weakly suppresses
  12. L7:2514 weakly suppresses
  13. L6:1319 weakly excites
  14. L6:1926 weakly excites
  15. L6:1391 weakly excites
  16. L0:438 weakly suppresses
  17. L7:2208 weakly excites
  18. L7:2057 weakly suppresses travel and journey verbs
  19. L7:583 weakly excites
  20. L7:348 weakly suppresses
  21. L6:2838 weakly excites
  22. L5:135 weakly excites
  23. L5:1765 weakly suppresses
  24. L4:1505 weakly suppresses

Feeds later layers

  1. L18:1478 suppresses
  2. L9:1699 suppresses
  3. L16:1121 suppresses
  4. L24:692 weakly suppresses
  5. L9:1460 weakly suppresses
  6. L17:2091 weakly suppresses
  7. L14:2989 weakly suppresses
  8. L11:1508 weakly excites
  9. L16:2175 weakly suppresses
  10. L13:1122 weakly suppresses
  11. L21:1544 weakly suppresses
  12. L13:2734 weakly excites
  13. L10:2145 weakly excites
  14. L9:1010 weakly suppresses
  15. L9:1995 weakly suppresses
  16. L10:1455 weakly suppresses
  17. L9:2665 weakly suppresses
  18. L14:1599 weakly suppresses
  19. L13:238 weakly excites
  20. L13:1380 weakly suppresses
  21. L25:2716 weakly suppresses
  22. L17:2675 weakly suppresses
  23. L17:2644 weakly excites
  24. L10:2791 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 -6.81 and -6.611 values between -3.65 and -3.451 values between -3.45 and -3.261 values between -2.86 and -2.671 values between -2.07 and -1.881 values between -1.68 and -1.481 values between -1.29 and -1.091 values between -1.09 and -0.894 values between -0.89 and -0.6910 values between -0.69 and -0.5042 values between -0.50 and -0.30598 values between -0.30 and -0.102303 values between -0.10 and 0.09821 values between 0.09 and 0.29101 values between 0.29 and 0.4916 values between 0.49 and 0.693 values between 0.69 and 0.882 values between 0.88 and 1.08offnegativepositiveoften

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. いっぱい
  2. anghai
  3. geçir
  4. ��
  5. rance
  6. cję
  7. yclopedia
  8. "`⏎⏎
  9. longitude
  10. _order

Suppresses

  1. ites
  2. 简单
  3. itz
  4. 金山
  5. basics
  6. fer
  7. simple
  8. 比较简单
  9. 简单的
  10. cts