Neuron Explorer

Qwen3-0.6B

comma in three-item lists

A neuron in layer 15 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 comma inside a list of three things, like 'existential, social, and moral' or 'red, green, and blue'. The steered text kept falling into list-like patterns of items, which fits that job.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    intellectual life more effectively.⏎Charlie Kurth, assistant professor in philosophy at Washington University in St Louis, Missouri, divides anxiety into three different categories: existential, social, and moral. The first, existential anxiety, is what we experience when we think

  2. #2 peak

    , but not to write settings to those files.⏎This topic describes the syntax of configuration files and provides information about the three types of configuration files: machine, application, and security.⏎Configuration files contain elements, which are logical data structures that set configuration

  3. #3 peak

    iparticle. With respect to symmetries and invariance principles, three special conservation laws have been described, associated with inversion or reversal of space, time, and charge.⏎Conservation laws are considered to be fundamental laws of nature, with

  4. #4 peak

    essential in determining the best choice of production strategy. Direct dynamic information from a reservoir may be obtained, in principle, from three sources: production history, pressure testing, and tracer testing. The value and importance of tracer tests are broadly recognized. Tr

  5. #5 peak

    address what’s causing it.⏎On a cellular level, brain fog is believed to be associated with high levels inflammation and changes to three primary hormones: dopamine, serotonin, and cortisol. One theory behind the underlying reason for brain fog symptoms is that higher

  6. #6 peak

    strikes. And they can dazzle the world with their cute googly eyes.⏎The world looks colorful to our eyes with only three color receptors — red, green, and blue. But for a mantis shrimp, a small marine crustacean

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 interest in the Chinese characters, the characters, and the characters' cultural significance, and the characters' historical, linguistic, and the characters' cultural diversity, the characters' linguistic, and the characters' cultural, linguistic, and the characters

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

    and the other is an antiparticle. With respect to symmetries and invariance principles, three special conservation laws have been described, associated with inversion or reversal of space, time, and charge.⏎Conservation laws are considered to be

  2. #2 peak

    78 and 2007, and tackled three different measures of income mobility: absolute mobility (as measured by inflation-adjusted income growth over time), and relative upward and downward mobility — i.e. movement up or down the socio-economic

  3. #3 peak

    ai and Ayutthaya, and was incorporated into the Kingdom of Siam in 1786. Patani was divided into the three provinces Patani, Yala, and Narathiwat in 1808.

  4. #4 peak

    a Deaf individual to discuss an interview, consider how phrases such as “oral history” or<|endoftext|>Everyone is familiar with three of the traditional states of matter (liquid, gas and solid) but the fourth fundamental state is one that is less well

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. L12:2723 excites
  2. L11:2760 suppresses
  3. L13:2022 excites
  4. L14:683 excites
  5. L13:1595 weakly suppresses
  6. L8:2675 weakly suppresses
  7. L11:2496 weakly suppresses
  8. L14:449 weakly suppresses
  9. L9:1636 weakly suppresses
  10. L11:1999 weakly excites
  11. L13:1258 weakly excites
  12. L2:2964 weakly suppresses
  13. L13:1389 weakly suppresses
  14. L10:2237 weakly excites
  15. L3:946 weakly suppresses
  16. L9:2696 weakly excites
  17. L8:1485 weakly excites
  18. L2:2190 weakly excites
  19. L11:2638 weakly excites
  20. L0:874 weakly excites
  21. L11:1178 weakly excites
  22. L5:1275 weakly suppresses
  23. L5:37 weakly excites
  24. L11:2248 weakly excites

Feeds later layers

  1. L16:2200 strongly excites
  2. L16:2360 strongly suppresses
  3. L19:492 strongly excites
  4. L20:882 strongly suppresses
  5. L16:2442 excites
  6. L17:1935 suppresses
  7. L17:1365 excites
  8. L18:2446 suppresses
  9. L16:1809 suppresses comma inside a list of names or acronyms
  10. L19:2552 suppresses
  11. L17:919 suppresses
  12. L18:1693 suppresses
  13. L16:1145 excites commas inside short lists
  14. L19:867 suppresses
  15. L16:2069 excites
  16. L20:1009 excites
  17. L20:75 suppresses
  18. L17:2610 excites and or comma closing a list
  19. L21:393 suppresses unclear: relative 'that' and list items
  20. L19:1089 suppresses
  21. L18:486 excites
  22. L17:1385 excites
  23. L16:1726 excites
  24. L19:2186 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 -1.32 and -1.222 values between -1.22 and -1.113 values between -1.11 and -1.012 values between -1.01 and -0.917 values between -0.91 and -0.8010 values between -0.80 and -0.7014 values between -0.70 and -0.6020 values between -0.60 and -0.5036 values between -0.50 and -0.3985 values between -0.39 and -0.29191 values between -0.29 and -0.19463 values between -0.19 and -0.08938 values between -0.08 and 0.021045 values between 0.02 and 0.12639 values between 0.12 and 0.23307 values between 0.23 and 0.33110 values between 0.33 and 0.4323 values between 0.43 and 0.543 values between 0.54 and 0.643 values between 0.64 and 0.742 values between 0.74 and 0.851 values between 0.85 and 0.951 values between 0.95 and 1.051 values between 1.46 and 1.571 values between 2.70 and 2.80offnegativepositiveoften

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. подпис
  2. /footer
  3. moderately
  4. Legend
  5. inding
  6. Investment
  7. invest
  8. anken
  9. lycer
  10. PEND

Suppresses

  1. рок
  2. orc
  3. 三层
  4. 三点
  5. .contentSize
  6. urt
  7. ican
  8. ndon
  9. 新三
  10. 四