Neuron Explorer

Qwen3-0.6B

second half of 'side by side' and 'face-to-face'

A neuron in layer 2 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 lights up on the last word of the phrases 'side by side' and 'face-to-face'. When it was turned up, the model's writing did not change in any clear way.

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

    , but also will give comprehensive information about condoms, contraception, and the prevention of sexually transmitted infections and diseases. When you put the two side by side, it seems like a clear-cut decision: abstinence-only curriculum will only further

  2. #2 peak

    with proper English wives. Simpson's former wife of ten years, Margaret Taylor, and Matooskie were unceremoniously dumped without even a face-to-face meeting. To his credit, Simpson made arrangements for the support of his children and

  3. #3 peak

    have to separate first? So that each homologue is on top of each other [on top - I mean from a birds eye view] instead of side by side? Are they always side by side when the cell is not replicating?⏎Would

  4. #4 peak

    This is a sometimes hypothetical pre-shoot narrative document describing what happens in the film from start to finish⏎- A Split Script – A side-by-side two column format with one side telling what the audience will see and the other side

  5. #5 peak

    zag across it, "very much like⏎surfing on a<|endoftext|>The power of art is sometimes found in the opposing juxtaposition of different images placed side by side. Artists place the delicate with the blunt, juxtapose complex with simple, or

  6. #6 peak

    Limited Interaction⏎The lack of interaction between teachers and students impacts learning in many ways. If a child is struggling with a particular concept or activity, a face-to-face interaction would help quickly identify and remedy the problem. An online instructor cannot see when

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 have been thinking about this for a long time, and I have been thinking about this for a long time. So I'm not sure of this. I am not sure of this. I'm not sure of this. I'm not

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

    active-causative הִפְעִיל verb form, in ביון, the root is plugged into the active-intensive פִּעֵל form yielding the verb לְבַ

  2. #2 peak

    ably in the Hebrew name of The Jewish Agency – הַסּוֹכְנוּת הַיְּהוּדִית, or simply הסוכנות-

  3. #3 peak

    ) is used; it indicates that God would make Pharaoh’s will strong or firm.⏎2 tn The form is וְאִכָּב

  4. #4 peak

    ence, and made, instead of Apap, the serpent of darkness, the great antagonist of the beneficent Osiris

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. L1:654 strongly suppresses
  2. L0:2639 strongly excites unclear: weak hits on 'holiday' and 'playing', negative on 'Ada-fruit'
  3. L0:2672 excites
  4. L1:767 suppresses
  5. L0:2518 suppresses
  6. L0:1780 suppresses
  7. L0:2583 suppresses
  8. L0:2415 suppresses
  9. L1:626 suppresses
  10. L0:2293 excites
  11. L1:910 excites
  12. L0:2785 suppresses
  13. L1:2278 excites
  14. L0:2416 excites
  15. L0:2844 excites
  16. L0:2582 suppresses
  17. L0:1947 suppresses
  18. L0:2857 suppresses
  19. L0:2972 excites
  20. L0:861 suppresses
  21. L0:2440 suppresses
  22. L1:1016 suppresses
  23. L0:2242 excites
  24. L0:2679 excites

Feeds later layers

  1. L3:2074 excites
  2. L3:849 suppresses
  3. L3:2874 suppresses
  4. L13:1468 weakly excites
  5. L20:1233 weakly excites
  6. L13:1853 weakly excites
  7. L17:1064 weakly excites
  8. L3:1760 weakly suppresses
  9. L15:1444 weakly suppresses
  10. L4:308 weakly excites
  11. L13:2794 weakly suppresses
  12. L16:2076 weakly excites
  13. L3:81 weakly suppresses
  14. L14:2622 weakly excites
  15. L12:1957 weakly excites
  16. L15:2400 weakly suppresses
  17. L9:266 weakly suppresses
  18. L3:2375 weakly excites
  19. L14:1567 weakly suppresses
  20. L10:1583 weakly excites
  21. L8:1205 weakly excites
  22. L13:1298 weakly excites
  23. L20:2018 weakly excites
  24. L17:845 weakly excites comma after an introductory phrase

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.

4 values between -0.52 and -0.3934 values between -0.39 and -0.27191 values between -0.27 and -0.141379 values between -0.14 and -0.022216 values between -0.02 and 0.1175 values between 0.11 and 0.235 values between 0.23 and 0.351 values between 0.35 and 0.481 values between 0.73 and 0.851 values between 3.96 and 4.081 values between 4.33 and 4.46offnegativepositiveoften

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. InSection
  2. /umd
  3. ня
  4. nym
  5. iya
  6. ями
  7. axed
  8. ость
  9. ости
  10. 网小编

Suppresses

  1. umeric
  2. 不吃
  3. iology
  4. repeating
  5. 回合
  6. timestamps
  7. Pre
  8. Boolean
  9. ographical
  10. 岁以下