Neuron Explorer

Qwen3-0.6B

unclear: relative 'that' and list items

A neuron in layer 21 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

The strongest responses are weak and point at words like 'that' starting a clause, but the negative side is much stronger and fires on items inside a list, like 'servers' in a list of cloud services. The overall idea is not clear. Turning it down made the model produce strings of commas and 'which', and turning it up gave generic sentences.

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

    this blog is not intended to be a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of qualified health providers with questions you may have regarding medical conditions.<|endoftext|>The Amendment Process and the African American

  2. #2 peak

    because they have higher kilojoule needs due to ongoing growth and development; women<|endoftext|>There are many cases in which you need a 2 dimensional lookup. That means, if you want to get a value from a specific row-column combination with

  3. #3 peak

    money mistakes I made as an early twenty-something. I want better for my kids. Here are 10 ways to teach financial literacy to your tween (that I wish someone taught me).⏎#1. Talk to Them About Financial Literacy

  4. #4 peak

    can build strength in your muscles, increase the amount of lean muscle in your body, manage your weight more easily and lower your risk of injury.⏎Examples of Weight-Training Exercises⏎You can perform weight-training exercises with free weights -- dumbbells

  5. #5 peak

    , recognize creatures and establish the parts of a vehicle, and a lot more.⏎You are able to customize them to meet up with your child’s specific requirements. That is amongst the most important good thing about the numbers worksheets. This can be accomplished

  6. #6 peak

    Click to take a picture of your objects. 4. Click then type more on the object that has the bigger capacity. 5. Click then type less on the object that

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, the, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I,

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

    far better thinker and<|endoftext|>Take any point P inside an equilateral triangle. Draw PA, PB and PC⏎from P perpendicular to the sides of the triangle where A, B and C⏎are points on the sides. Prove that PA +

  2. #2 peak

    , cloud computing is a technology used for storing and accessing data, application and programs over the Internet. Cloud computing refers to the delivery of hosted and computing services including servers, storage, databases, networking, analytics, and intelligence over

  3. #3 peak

    ventral (y) and rostro-caudal (z). Another one is a stereotactic device or apparatus which uses a set of three coordinates (x, y and z). And the last one is stereotactic localization and placement

  4. #4 peak

    male, based on the warty pig Species Survival Plan coordinator’s recommendation to improve the<|endoftext|>Learning the Relationship Between Culture and Relationships⏎Culture is the total set of beliefs, values, manners and customs that are discovered and distributed by

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. L19:1089 strongly excites
  2. L20:1083 strongly suppresses
  3. L19:1394 strongly excites
  4. L20:610 strongly suppresses
  5. L19:3066 strongly excites
  6. L19:2704 strongly suppresses
  7. L18:1508 strongly suppresses
  8. L19:2835 suppresses
  9. L19:492 suppresses
  10. L20:1744 suppresses comma after a sentence's subject
  11. L18:120 suppresses
  12. L20:2864 suppresses
  13. L19:1813 suppresses
  14. L19:164 suppresses
  15. L20:822 suppresses
  16. L15:132 suppresses
  17. L20:2865 suppresses
  18. L19:2552 excites
  19. L17:2323 excites
  20. L19:290 excites
  21. L19:1510 excites
  22. L15:2999 excites
  23. L20:132 suppresses
  24. L18:957 excites

Feeds later layers

  1. L26:236 strongly suppresses
  2. L27:86 strongly excites
  3. L26:22 strongly excites the 'res' fragment inside words
  4. L22:368 strongly suppresses
  5. L24:216 strongly suppresses
  6. L25:438 strongly suppresses
  7. L27:286 strongly suppresses
  8. L25:328 strongly suppresses
  9. L27:358 strongly suppresses
  10. L27:191 strongly suppresses
  11. L22:940 strongly suppresses
  12. L25:1221 strongly suppresses
  13. L24:572 strongly suppresses
  14. L26:80 strongly excites
  15. L26:307 excites
  16. L25:969 excites
  17. L27:130 excites
  18. L24:304 suppresses
  19. L22:158 excites
  20. L24:6 excites
  21. L26:345 suppresses
  22. L24:32 suppresses
  23. L25:346 excites
  24. L27:291 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 -28.95 and -28.091 values between -24.65 and -23.791 values between -23.79 and -22.932 values between -22.07 and -21.213 values between -21.21 and -20.352 values between -19.49 and -18.633 values between -17.77 and -16.911 values between -16.05 and -15.195 values between -15.19 and -14.333 values between -14.33 and -13.475 values between -13.47 and -12.612 values between -12.61 and -11.755 values between -11.75 and -10.897 values between -10.89 and -10.034 values between -10.03 and -9.1710 values between -9.17 and -8.3114 values between -8.31 and -7.456 values between -7.45 and -6.5917 values between -6.59 and -5.7320 values between -5.73 and -4.8724 values between -4.87 and -4.0145 values between -4.01 and -3.1580 values between -3.15 and -2.29136 values between -2.29 and -1.43362 values between -1.43 and -0.571950 values between -0.57 and 0.29926 values between 0.29 and 1.15215 values between 1.15 and 2.0147 values between 2.01 and 2.876 values between 2.87 and 3.733 values between 3.73 and 4.592 values between 4.59 and 5.45offnegativepositiveoften

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. 、“
  2. "),"
  3. ）、
  4. '),('
  5. 、、
  6. 😉⏎⏎
  7. """),⏎
  8. ;-)
  9. ";//
  10. ;m

Suppresses

  1. ,
  2. ，
  3. (),
  4. ,h
  5. ،
  6. ,
  7. ,\
  8. ，在
  9. ++,
  10. ,',