Neuron Explorer

Qwen3-0.6B

unclear: 'as' after a verb phrase

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

This neuron is hard to pin down. It fires most on 'as' after verbs like 'emerges' or 'slashed', and it dips on 'the' in phrases like 'off the mark'. Turning it up gave off-topic text, so the steering does not explain it.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    HEXA-1,3,5-TRIENE). AROMATIC COMPOUNDS ARE ALWAYS FOLLOWS THE SUBSTITUTION REACTION BECAUSE OF THE STABILITY OF THE BENZENE RING. IT WILL

  2. #2 peak

    human colon cancer cell-lines has demonstrated that the “fat hormone” leptin may enhance the growth of colonic cancer cells. The discovery not only offers an explanation as to the underlying cause of the increase colon cancer risk in obesity but could lead to

  3. #3 peak

    of “students like you, who found concepts such as this difficult, tended to do better in tests when they used these learning materials”.⏎While the data generated as a byproduct of online courses provide a valuable resource in their own right<|endoftext|>The

  4. #4 peak

    in reverse order? According to the laws of physics, this can happen.⏎NARRATOR: It's a world that comes to light as we probe the most extreme realms of the cosmos, from black holes to the Big

  5. #5 peak

    students to report<|endoftext|>Farmers in designated Nitrate Vulnerable Zones (NVZs) are to have the amount of manure they can spread on fields slashed as Defra seeks to combat the problems caused by over-fertilisation.⏎The ‘

  6. #6 peak

    on the size confinement. By analyzing the growth temperature dependence, we unambiguously revealed that there is a coupling between the LLCCs and the inner tubes as strongly supported by a model of coupled Raman modes. This allows us to directly

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 world of a lot of people, the world of a lot of people, the world of a lot of people, the world of a lot of people, the world of a lot of people, the world of a lot of people,

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

    page of the newspaper on this day in 1933. Hitler was made Chancellor on January 30th. So, what was he up to the day after the Reichstag fire, as published on March 1st?⏎-

  2. #2 peak

    , 2001: A Space Odyssey, Stargate, and Avatar. As is often the case, sci-fi may not be that far off the mark, as NASA believes that stasis will be crucial to future missions to Mars

  3. #3 peak

    cadmium in children’s products– today the agency backed down. They are suggesting that the industry self-enforce. Hmmmm? How well did that work out the last time when we were talking about lead. We are especially concerned that children’s

  4. #4 peak

    Types of Sources for a Research Paper | Pen and the Pad There are two types of research<|endoftext|>The Apollo 11 space flight landed the first humans on Earth’s Moon on July 20 1969. The mission, carried

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. L6:65 excites
  2. L5:227 suppresses
  3. L5:33 excites
  4. L5:1459 excites
  5. L5:2566 excites
  6. L5:38 suppresses
  7. L6:158 excites
  8. L1:1192 weakly excites
  9. L4:78 weakly suppresses line breaks inside a sentence
  10. L3:123 weakly excites
  11. L6:10 weakly excites
  12. L4:2901 weakly suppresses
  13. L5:3 weakly excites
  14. L4:84 weakly excites
  15. L6:161 weakly suppresses
  16. L2:501 weakly excites
  17. L5:955 weakly excites
  18. L4:632 weakly suppresses
  19. L5:2685 weakly excites
  20. L4:75 weakly suppresses
  21. L5:129 weakly excites
  22. L6:924 weakly suppresses
  23. L1:739 weakly excites
  24. L3:927 weakly suppresses

Feeds later layers

  1. L8:382 suppresses
  2. L8:967 suppresses
  3. L14:375 suppresses
  4. L11:2632 suppresses
  5. L15:2081 suppresses
  6. L10:1985 suppresses
  7. L16:1540 excites
  8. L8:1433 suppresses
  9. L15:1495 suppresses
  10. L8:781 suppresses
  11. L15:1653 suppresses
  12. L15:817 suppresses
  13. L15:540 weakly excites
  14. L15:2713 weakly suppresses
  15. L9:268 weakly suppresses
  16. L8:411 weakly suppresses
  17. L17:798 weakly suppresses
  18. L8:602 weakly suppresses
  19. L15:762 weakly suppresses
  20. L20:423 weakly excites
  21. L16:247 weakly suppresses
  22. L16:1475 weakly suppresses
  23. L8:326 weakly excites
  24. L15:482 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 -2.73 and -2.622 values between -1.82 and -1.713 values between -1.71 and -1.601 values between -1.60 and -1.481 values between -1.37 and -1.262 values between -1.26 and -1.143 values between -1.14 and -1.033 values between -1.03 and -0.926 values between -0.92 and -0.8013 values between -0.80 and -0.6916 values between -0.69 and -0.5813 values between -0.58 and -0.4645 values between -0.46 and -0.35113 values between -0.35 and -0.24268 values between -0.24 and -0.12997 values between -0.12 and -0.011646 values between -0.01 and 0.11488 values between 0.11 and 0.22158 values between 0.22 and 0.3354 values between 0.33 and 0.4535 values between 0.45 and 0.5616 values between 0.56 and 0.6713 values between 0.67 and 0.794 values between 0.79 and 0.902 values between 0.90 and 1.012 values between 1.01 and 1.131 values between 1.13 and 1.241 values between 1.35 and 1.471 values between 1.70 and 1.81offnegativepositiveoften

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. 如果是
  2. 若您
  3. pu
  4. を求め
  5. 若
  6. 若有
  7. 其中
  8. geometric
  9. isError
  10. 其中一个

Suppresses

  1. -level
  2. anding
  3. hood
  4. ansa
  5. 렉
  6. 级
  7. izr
  8. ional
  9. ળ
  10. -Level