Neuron Explorer

Qwen3-0.6B

'so' in 'and so on'

A neuron in layer 18 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' in phrases like 'and so on' or 'and so were'. Pushing it up made the model repeat the word 'then' over and over, because the neuron tends to push that word forward. Its negative side is on sentence starts like 'Now' and 'Then' that begin a new step.

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

    chemical reaction and so can be reused.⏎- Enzymes are proteins and so are sensitive to temperature

  2. #2 peak

    Cold War. That was before scientists knew about the oceans on moons around other planets, before they knew about how tough microorganisms get here on Earth (and so maybe in space too?), before they started planning experiments to look for life on Mars

  3. #3 peak

    in terms of safety, reliability, performance, profitability, and credibility for the affected organization and its representatives.⏎There are standards for quality, safety, reliability, and so on, in industries related to food, drugs, military equipment, manufacturing, computer

  4. #4 peak

    it to their meaningful contexts, finding it in interesting media and so on. Only when love is strong, memory work can follow.⏎Yesterday, we

  5. #5 peak

    which you don’t know the return column. For example, depending on the specifications, you want to return the value from either column B, C or D and so on. In such case, we are speaking of 2D lookups or

  6. #6 peak

    but so is the rest of the Middle East. In Egypt, potential presidential contender Gamal

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: if the question, as stated in the question, says that the problem says that the problem says that, for example, for example, in a problem that is said in the statement, the problem says that, in the problem statement, a

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

    , mull and the covering material as depicted in figure 180. While doing this, make sure that the blade does not reach the first signature. Now cut off the threads as shown in figure 181 and

  2. #2 peak

    Saint-Galmier (Loire).⏎Seven years later, he found a partnership with Bessac and the painter Joséphus Veyrat. Then, he set up his own business in Grigny, around ten kilometres from

  3. #3 peak

    at some distant object like a mountain or tree, hold your thumb at arm’s length, close one eye, and then move your thumb to block the object. Now, holding your thumb still, move your head in a small circle. Even though

  4. #4 peak

    : improving our health and protecting the environment.⏎We’ve addressed the health benefits of choosing a plant-based diet in our blog ‘8 Health Benefits of Going Vegan’. Now let’s take a look at the benefits of a vegan diet for our environment…⏎

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. L17:845 excites comma after an introductory phrase
  2. L17:1716 excites
  3. L11:1370 weakly excites 'then' in step-by-step sequences
  4. L17:1067 weakly excites
  5. L17:1426 weakly excites
  6. L3:1914 weakly suppresses
  7. L4:42 weakly suppresses
  8. L10:341 weakly excites "of" inside treaty names
  9. L13:1108 weakly suppresses
  10. L16:77 weakly excites
  11. L2:1494 weakly suppresses
  12. L11:31 weakly suppresses
  13. L2:1451 weakly suppresses
  14. L5:1755 weakly excites
  15. L5:129 weakly excites
  16. L3:859 weakly suppresses
  17. L6:330 weakly excites
  18. L6:84 weakly excites
  19. L2:205 weakly suppresses
  20. L12:2911 weakly suppresses
  21. L4:592 weakly excites
  22. L14:2475 weakly suppresses
  23. L0:1815 weakly excites
  24. L8:705 weakly excites

Feeds later layers

  1. L23:221 excites
  2. L27:1006 excites
  3. L20:2429 excites
  4. L20:663 excites
  5. L22:1056 excites
  6. L25:1425 excites
  7. L19:1593 excites
  8. L19:2501 weakly suppresses
  9. L19:2110 weakly excites
  10. L21:1546 weakly excites
  11. L23:3040 weakly excites
  12. L19:934 weakly excites
  13. L27:1964 weakly excites
  14. L21:2083 weakly excites
  15. L26:1007 weakly excites
  16. L24:423 weakly excites
  17. L25:2171 weakly suppresses
  18. L20:513 weakly excites
  19. L19:1128 weakly excites
  20. L20:1811 weakly excites 'by' before a deadline
  21. L23:621 weakly excites
  22. L23:2377 weakly excites
  23. L27:1750 weakly excites
  24. L26:304 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 -14.81 and -14.271 values between -13.73 and -13.191 values between -12.66 and -12.121 values between -8.89 and -8.361 values between -7.28 and -6.741 values between -6.74 and -6.212 values between -6.21 and -5.671 values between -5.67 and -5.131 values between -5.13 and -4.593 values between -4.59 and -4.064 values between -4.06 and -3.526 values between -3.52 and -2.9810 values between -2.98 and -2.449 values between -2.44 and -1.9123 values between -1.91 and -1.3745 values between -1.37 and -0.83203 values between -0.83 and -0.292831 values between -0.29 and 0.25697 values between 0.25 and 0.7826 values between 0.78 and 1.3210 values between 1.32 and 1.868 values between 1.86 and 2.408 values between 2.40 and 2.934 values between 2.93 and 3.473 values between 3.47 and 4.014 values between 4.01 and 4.552 values between 5.62 and 6.162 values between 6.16 and 6.70offnegativepositiveoften

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. then
  2. then
  3. Then
  4. Then
  5. 然后
  6. →then
  7. 然後
  8. THEN
  9. THEN
  10. 然后再

Suppresses

  1. came
  2. comes
  3. icular
  4. store
  5. くる
  6. .setProperty
  7. prü
  8. VarChar
  9. filer
  10. 毫