Neuron Explorer

Qwen3-0.6B

'as' in 'seems as if' comparisons

A neuron in layer 20 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 lights up on the word 'as' in phrases like 'it seems as if' or 'looked as though', where someone is describing how something appears. When it was turned up, the model fell into repeating 'if' and 'which of the following', so the link to the pattern is unclear.

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

    his book:⏎The two entities that were supposed to write the regulations, the Pay Board and the Price Commission, wrangled and dithered. When it looked as though they were going to miss a crucial deadline

  2. #2 peak

    chairman Mahmoud Abbas is accusing Israel of “genocide,” and the PA’s newspaper is calling the war “Israel’s Holocaust.”⏎It seems as if every time Israel defends itself, somebody points an accusing finger and yells “

  3. #3 peak

    y⏎May 27, 2012⏎When the last American astronauts blasted off from the Moon in 1972, it seemed as if they were leaving behind monuments that would stand for all time. On a life

  4. #4 peak

    Frankenstein’s monster, consisting of two identical M1 Max chips bolted together using a silicon interface that serves as a bridge. This clever design makes it seem as if the conjoined chips are in fact just

  5. #5 peak

    (the form is at the bottom of the page).⏎Esther – Day 3⏎Xerxes wanted to show off his beautiful wife! But it seems as though he did not give her too much thought until he had drunk too much!

  6. #6 peak

    for she was lovely to look at.⏎As the story continues, pride creeps in.⏎Xerxes wanted to show off his beautiful wife! But it seems as though he didn’t give her too much thought until he had drunk too much!⏎

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 first time the movie was released, the main plot the first movie the movie was talking time the time the movie going going compared the time the movie going the time the movie talking The first movie the movie was talking time the time the

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

    in this world?⏎Close MATLAB with keyboard⏎Find the treasures in MATLAB Central and discover how the community can help you!⏎Choose a web site to get translated content where available and see local events and offers.

  2. #2 peak

    Interior Angle Sum⏎Bit to Gigabyte⏎Find the treasures in MATLAB Central and discover how the community can help you!⏎Choose a web site to get translated content where available and

  3. #3 peak

    well-being for individuals with mental health conditions. This publication provides a snapshot of existing instruments that measure various domains of wellness and well-being, as well as physical health where it intersects with these domains. This resource emphasizes non-proprietary, public-access

  4. #4 peak

    making an orthosis involves several steps:1⏎- An assessment of the patient’s medical condition and needs, including diagnosis, pressure mapping, and gait analysis where indicated.⏎- The creation of a positive cast or digital image of the foot,

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. L19:1815 suppresses
  2. L19:1703 weakly excites
  3. L19:1790 weakly suppresses
  4. L16:180 weakly excites
  5. L19:1406 weakly suppresses
  6. L18:238 weakly suppresses
  7. L9:523 weakly suppresses
  8. L16:169 weakly excites
  9. L16:2532 weakly suppresses
  10. L5:2280 weakly excites
  11. L8:7 weakly excites
  12. L17:137 weakly excites
  13. L16:209 weakly excites
  14. L16:598 weakly excites
  15. L4:2904 weakly excites
  16. L14:196 weakly excites
  17. L7:2869 weakly excites
  18. L0:860 weakly excites chinese 的 particle before a noun
  19. L19:2531 weakly excites
  20. L17:56 weakly excites
  21. L19:16 weakly suppresses
  22. L0:1085 weakly excites
  23. L4:2975 weakly excites
  24. L18:739 weakly suppresses

Feeds later layers

  1. L25:2663 excites
  2. L24:148 excites
  3. L22:427 weakly excites
  4. L27:1084 weakly excites
  5. L27:338 weakly excites
  6. L25:1503 weakly excites
  7. L21:1746 weakly excites
  8. L23:449 weakly excites
  9. L21:1064 weakly excites
  10. L25:2548 weakly excites
  11. L24:422 weakly suppresses
  12. L26:466 weakly suppresses
  13. L23:688 weakly excites
  14. L21:793 weakly excites
  15. L24:557 weakly excites
  16. L22:160 weakly suppresses
  17. L23:1172 weakly excites
  18. L25:147 weakly excites
  19. L26:1021 weakly excites
  20. L21:1814 weakly suppresses
  21. L22:1720 weakly excites
  22. L21:110 weakly excites
  23. L24:100 weakly suppresses
  24. L21:1931 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 -38.84 and -37.592 values between -28.82 and -27.571 values between -27.57 and -26.324 values between -26.32 and -25.071 values between -25.07 and -23.813 values between -22.56 and -21.314 values between -18.80 and -17.554 values between -16.30 and -15.052 values between -15.05 and -13.791 values between -13.79 and -12.541 values between -12.54 and -11.292 values between -11.29 and -10.042 values between -10.04 and -8.784 values between -8.78 and -7.533 values between -7.53 and -6.288 values between -6.28 and -5.0313 values between -5.03 and -3.7726 values between -3.77 and -2.5292 values between -2.52 and -1.271241 values between -1.27 and -0.022415 values between -0.02 and 1.2448 values between 1.24 and 2.4920 values between 2.49 and 3.745 values between 3.74 and 5.004 values between 5.00 and 6.251 values between 10.01 and 11.26offnegativepositiveoften

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. ually
  2. though
  3. though
  4. erator
  5. although
  6. lessly
  7. iously
  8. 因为它
  9. edly
  10. енно

Suppresses

  1. applicable
  2. ob
  3. esper
  4. ver
  5. demonstrated
  6. apro
  7. доступ
  8. well
  9. asked
  10. available