Neuron Explorer

Qwen3-0.6B

'and' joining clauses in instructions

A neuron in layer 11 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 switches on at the word 'and' when it links two parts of a sentence, especially in instructions or rules, such as 'unless you're a fatso and don't move' or 'Join our program and our community'. Turning it up did not produce a clear theme; the continuations wandered off into unrelated topics.

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

    them every day (unless you're a fatso and don't move much or use a lift each time you need to get onto a

  2. #2 peak

    1853–1930⏎Our editors will review what you’ve submitted and determine whether to revise the article.Join Britannica's Publishing Partner Program and our community of experts to gain a global audience for your work!⏎One of the

  3. #3 peak

    : no pets, carry out all your trash when you leave, be out by nightfall unless someone is hurt and waiting for assistance, no camping, and if you encounter indigenous artifacts in the caves

  4. #4 peak

    ⏎- Foods that contain the following ingredients: questionable flavours, emulsifiers, enzymes, and gelatine⏎- Pig and dog⏎- The by-product of the human

  5. #5 peak

    in the streets of the capital.⏎The contents of this article is licensed from www.wikipedia.org under the GNU Free Documentation License. Click here to see the transparent copy and copyright details<|endoftext|>What is Celiac / Coeliac Disease?Tweet⏎What is

  6. #6 peak

    jumper result in 3 different Transmission curves when using⏎a Network Analyser or Spectrum Analyser with built in generator.⏎( of course you may use a generator and a hf-voltmeter or anything equivalent ... )⏎The Amplitude response looks somehow

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 don't have it. I don't have it. I think it's the first thing. I think it's the second thing. I think it's the first thing... and I just want to make a noise. I think it

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

    location of the monument. Where it is not available, the satellite view is shown instead.⏎Latitude: 52.5162 / 52°30'58"N⏎Longitude: -3.0604

  2. #2 peak

    ening qualities as well as a perceived belief that it possessed healing and cosmetic properties. The gum is still used today as a spice in cooking all over the world while also serving as a key ingredient in products like toothpaste, skin lotions, and

  3. #3 peak

    watch on YouTube. A prestigious panel of scientists, astronauts and educators which include His Royal Highness Prince Sultan bin Salman bin Abdul-Aziz Al Saud, astronaut and payload specialist aboard space shuttle mission STS-51-G in 19

  4. #4 peak

    survey of the history of science from its roots in different cultures around the world through to the present day. He explores crucial milestones in scientific development and at the same time examines the enormous social and intellectual changes they initiated. Opening with a discussion of the

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. L10:135 suppresses 'or' in alternative requirements
  2. L10:74 excites
  3. L4:57 suppresses
  4. L7:1475 suppresses 'or' joining alternatives in lists
  5. L9:92 suppresses "and" or "or" joining two items
  6. L8:70 excites 'and' between stacked superlatives
  7. L5:548 excites
  8. L10:205 excites
  9. L10:901 excites
  10. L3:1742 weakly suppresses
  11. L4:131 weakly suppresses
  12. L3:447 weakly excites
  13. L1:916 weakly excites
  14. L10:1643 weakly suppresses
  15. L10:2944 weakly excites
  16. L7:999 weakly suppresses
  17. L4:291 weakly excites
  18. L7:2416 weakly excites
  19. L7:1269 weakly excites
  20. L10:2466 weakly suppresses
  21. L5:560 weakly excites
  22. L8:1173 weakly excites
  23. L5:2912 weakly suppresses
  24. L4:2098 weakly suppresses

Feeds later layers

  1. L15:1907 suppresses the word 'or' between alternatives
  2. L12:1454 excites
  3. L12:1197 excites
  4. L15:2203 excites
  5. L15:2935 suppresses
  6. L13:553 suppresses
  7. L12:2279 excites
  8. L13:878 excites
  9. L12:1094 excites
  10. L13:2356 suppresses
  11. L18:2197 suppresses
  12. L14:1370 excites
  13. L17:2166 weakly suppresses
  14. L12:2965 weakly excites
  15. L16:376 weakly excites 'or' joining alternatives
  16. L16:935 weakly excites
  17. L12:2058 weakly excites
  18. L16:530 weakly suppresses
  19. L15:449 weakly excites
  20. L14:1213 weakly excites
  21. L13:887 weakly excites
  22. L20:1719 weakly excites
  23. L16:1913 weakly excites
  24. L12:1452 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 -1.24 and -1.121 values between -1.12 and -1.001 values between -1.00 and -0.882 values between -0.88 and -0.761 values between -0.76 and -0.6421 values between -0.64 and -0.5242 values between -0.52 and -0.39149 values between -0.39 and -0.27528 values between -0.27 and -0.151178 values between -0.15 and -0.031437 values between -0.03 and 0.09347 values between 0.09 and 0.2195 values between 0.21 and 0.3325 values between 0.33 and 0.4519 values between 0.45 and 0.5710 values between 0.57 and 0.6914 values between 0.69 and 0.815 values between 0.81 and 0.944 values between 0.94 and 1.065 values between 1.06 and 1.181 values between 1.18 and 1.304 values between 1.30 and 1.422 values between 1.42 and 1.544 values between 1.78 and 1.901 values between 1.90 and 2.021 values between 2.02 and 2.143 values between 2.26 and 2.381 values between 2.38 and 2.512 values between 2.51 and 2.631 values between 2.63 and 2.751 values between 2.75 and 2.871 values between 3.23 and 3.351 values between 3.47 and 3.59offnegativepositiveoften

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. zwar
  2. upal
  3. 铗
  4. /or
  5. )+(
  6. xac
  7. vows
  8. istringstream
  9. #undef
  10. ADM

Suppresses

  1. unner
  2. pseudo
  3. 这句话
  4. зер
  5. сет
  6. mere
  7. ibili
  8. ased
  9. verse
  10. ippo