Neuron Explorer

Qwen3-0.6B

'so' after a comma

A neuron in layer 12 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 word 'so' when it follows a comma and introduces a result, as in 'the area may be busy, so it's a good idea to let people know'. Its negative side is weak and shows on 'you' or 'I've' in site boilerplate. Turning it up did not make the output about cause and effect.

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

    practice for talking, so encourage your baby's efforts by chatting and reading together. Your baby is also figuring out how objects relate to each other and fit together, so it's a perfect time for stacking toys. This week, get tips on ways

  2. #2 peak

    a web search can help you locate the telephone numbers⏎and email addresses of contacts that may be able to help you.⏎Some stations may be on restricted property, so it’s a good idea⏎to let people know you are coming.⏎At the

  3. #3 peak

    to work on by using a noncontact voltage tester. In many cases, more than one circuit may run through a lighting outlet box or through a junction box, so you may have to turn off

  4. #4 peak

    , They protect your eyes from ultraviolet rays.⏎Keep in mind that your body needs a bit of healthy fat to best absorb these eye-benefitting nutrients, so try eating butternut squash with a little drizzle of olive oil.⏎It

  5. #5 peak

    best bet for finding personal news!) However, be aware that sometimes short notices (i.e. birth, marriage or death) are fit in between longer articles, so they could be anywhere!⏎Obituaries may be among the easier records to find

  6. #6 peak

    by relating the fictional events of a woman's life to a contemporary meditation on the meaning of the story of Lot's wife.⏎This article is available to subscribers only. Please subscribe for full access—subscriptions begin at $2.95. Already

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'm a student in the class of a math major, or a math major in the class of a math major, or a math major in the class of a math teacher, or a math teacher in the class of a math major

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

    OSC⏎ALL IAT SQU⏎See another brain teaser just like this one...⏎Or, just get a random brain teaser⏎If you become a registered user you can vote on this brain teaser, keep track of⏎which ones you have seen

  2. #2 peak

    comprehensive bundle of over 15 products for practically every Fine Motor Skills Alphabet Activity you can possibly think of! However, it’s a rapidly growing bundle and I’ve got even more ideas coming down the pike!⏎You’ll be surprised to find

  3. #3 peak

    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 you can vote on this brain teaser, keep track of⏎which ones you have seen

  4. #4 peak

    watching in real-time as the COVID-19 virus continues to change and become more transmissible.”⏎Kindly follow us on Twitter⏎You must be logged in to<|endoftext|>Children can be stubborn, willful, and infuriating.⏎But maybe that

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. L11:877 excites
  2. L4:365 weakly excites
  3. L11:675 weakly excites
  4. L10:2037 weakly excites
  5. L8:1949 weakly excites
  6. L7:1293 weakly excites
  7. L11:10 weakly excites
  8. L7:3065 weakly excites
  9. L8:1944 weakly suppresses
  10. L6:1764 weakly excites
  11. L10:2952 weakly suppresses
  12. L2:2391 weakly excites
  13. L10:844 weakly suppresses
  14. L6:2580 weakly excites
  15. L1:2274 weakly excites
  16. L7:856 weakly suppresses
  17. L1:1742 weakly excites
  18. L3:662 weakly suppresses
  19. L9:1242 weakly excites
  20. L11:328 weakly suppresses
  21. L6:2982 weakly excites
  22. L2:18 weakly excites
  23. L8:2104 weakly excites
  24. L8:102 weakly excites

Feeds later layers

  1. L14:2124 excites
  2. L17:2433 excites
  3. L19:730 suppresses
  4. L17:1283 excites
  5. L15:2624 excites
  6. L16:949 excites
  7. L20:2190 excites
  8. L22:1938 excites
  9. L18:2742 excites
  10. L18:2489 excites
  11. L13:845 excites
  12. L14:1244 excites
  13. L18:1862 excites
  14. L18:2125 weakly excites
  15. L21:1361 weakly excites
  16. L15:1574 weakly excites
  17. L15:641 weakly excites
  18. L18:2577 weakly excites
  19. L16:2563 weakly excites
  20. L19:3044 weakly excites
  21. L13:530 weakly excites commas after a 'since' clause
  22. L14:2568 weakly excites
  23. L15:2916 weakly excites
  24. L19:1824 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.

2 values between -2.05 and -1.941 values between -1.72 and -1.611 values between -1.50 and -1.401 values between -1.18 and -1.072 values between -0.96 and -0.857 values between -0.85 and -0.745 values between -0.74 and -0.6412 values between -0.64 and -0.5317 values between -0.53 and -0.4272 values between -0.42 and -0.31212 values between -0.31 and -0.20453 values between -0.20 and -0.091238 values between -0.09 and 0.021026 values between 0.02 and 0.13538 values between 0.13 and 0.24213 values between 0.24 and 0.3472 values between 0.34 and 0.4515 values between 0.45 and 0.566 values between 0.56 and 0.673 values between 0.67 and 0.782 values between 0.78 and 0.892 values between 0.89 and 1.003 values between 1.10 and 1.211 values between 1.21 and 1.321 values between 1.32 and 1.431 values between 1.43 and 1.541 values between 1.76 and 1.861 values between 2.19 and 2.30offnegativepositiveoften

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. hãy
  2. ぜひ
  3. ropol
  4. bitte
  5. 该怎么办
  6. .copyOf
  7. 要好好
  8. tent
  9. untas
  10. enic

Suppresses

  1. ######⏎
  2. _ng
  3. ivo
  4. phem
  5. PEnd
  6. alloween
  7. -purple
  8. 会长
  9. <?,
  10. 中午