Neuron Explorer

Qwen3-0.6B

help-seeking 'I don't know' phrases

A neuron in layer 8 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on phrases where someone admits they are stuck or do not know something, such as 'I'm lost', 'I don't know', or 'where to start'. Turning it up pushed the model toward writing about problems and solving them.

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

    is⏎I got the f(x) = x-1 part, and I got how to calculate the variance after you have the expected values, but I'm lost on other questions.⏎My questions are:⏎Where do we get if x=1

  2. #2 peak

    , Past, Future Indefinite, Present, Past, Future Continuous⏎помогите пожаааалуйста, сам не справляюсь(⏎Y. Раскройте скобки, употребляя г

  3. #3 peak

    3 (molar)⏎l is the path length in centimetres⏎i want to calculate the concentraton only after determining the extinction coeff. then hw do i determine the extinction coeff of supercoiled DNA, for example??⏎if you

  4. #4 peak

    is my weakness programming i feel an igonorant when it comes to programming my filed requires from me to take this course and i dont know anything,. sombody can help me with following program please!!:⏎Write an assembly

  5. #5 peak

    their asceticism, especially by their attachment to dhikr, the practice of remembrance of God, often performed after prayers<|endoftext|>Don’t know where to start with growing from seed? Are you worried that it will be difficult & require special

  6. #6 peak

    given to you?⏎October 3rd, 2010, 12:50 PM⏎i started the homework⏎i dont know how to finish it here what i

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 will be the one who follow the same pattern as the first problem. The problem is about the 2020 deadline 10 (10) (10) (10) (1) (11)

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

    want to calculate the concentraton only after determining the extinction coeff. then hw do i determine the extinction coeff of supercoiled DNA, for example??⏎if you've your three forms in the sample i

  2. #2 peak

    are the part of same Microsoft exchange server.⏎How do I know if a recalled communication was successful in outlook?⏎Make sure you have to tick the following option when recalling the email. Outlook will send you a

  3. #3 peak

    top percentage of the class, or when a student drops to the lowest percentages of the class. Select the stream alert to view the data.⏎In the list where your name appears, select Grades. On the Grades page, select the cell that displays

  4. #4 peak

    policy, in a simulated domain.⏎the application examples in programming are very good and i loved how rl gets closer and closer to how a living being thinks. if you take a course in audit mode, you will be able to see most course materials

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. L3:255 excites
  2. L3:148 suppresses
  3. L7:676 weakly suppresses
  4. L7:1076 weakly suppresses
  5. L7:61 weakly excites
  6. L7:30 weakly suppresses
  7. L6:1167 weakly suppresses
  8. L7:563 weakly excites
  9. L7:1981 weakly excites
  10. L2:1350 weakly suppresses
  11. L6:2189 weakly excites
  12. L1:599 weakly suppresses
  13. L1:2128 weakly suppresses
  14. L5:1466 weakly suppresses
  15. L4:413 weakly excites
  16. L6:17 weakly suppresses
  17. L6:16 weakly excites
  18. L3:166 weakly suppresses
  19. L6:637 weakly excites
  20. L6:2437 weakly suppresses
  21. L3:213 weakly suppresses
  22. L1:887 weakly excites 'be' after 'might', 'also' or 'only'
  23. L7:1698 weakly suppresses
  24. L1:566 weakly excites "addition" in "In addition"

Feeds later layers

  1. L9:373 excites
  2. L13:1309 suppresses
  3. L9:276 suppresses
  4. L9:2680 excites
  5. L9:655 weakly suppresses
  6. L11:2376 weakly excites
  7. L10:1112 weakly excites
  8. L10:2808 weakly excites
  9. L9:585 weakly excites
  10. L11:561 weakly suppresses
  11. L10:1342 weakly excites
  12. L14:28 weakly excites
  13. L12:863 weakly excites
  14. L11:2350 weakly excites
  15. L11:2791 weakly excites
  16. L11:2414 weakly suppresses
  17. L13:888 weakly excites
  18. L10:3048 weakly suppresses
  19. L9:184 weakly suppresses
  20. L9:757 weakly suppresses
  21. L19:1547 weakly suppresses
  22. L9:1207 weakly excites unclear: mid-sentence tokens in web text
  23. L14:1218 weakly suppresses
  24. L20:2629 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.

2 values between -1.60 and -1.491 values between -1.49 and -1.392 values between -1.18 and -1.082 values between -1.08 and -0.972 values between -0.97 and -0.871 values between -0.87 and -0.773 values between -0.77 and -0.663 values between -0.66 and -0.567 values between -0.56 and -0.4613 values between -0.46 and -0.3565 values between -0.35 and -0.25278 values between -0.25 and -0.15730 values between -0.15 and -0.041467 values between -0.04 and 0.06825 values between 0.06 and 0.17324 values between 0.17 and 0.27132 values between 0.27 and 0.3734 values between 0.37 and 0.488 values between 0.48 and 0.582 values between 0.58 and 0.681 values between 0.79 and 0.891 values between 1.10 and 1.201 values between 1.20 and 1.301 values between 1.30 and 1.411 values between 1.61 and 1.721 values between 2.34 and 2.441 values between 2.44 and 2.55offnegativepositiveoften

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. يد
  2. roid
  3. factors
  4. typeId
  5. オス
  6. ueil
  7. leshoot
  8. /sources
  9. 装修
  10. isors

Suppresses

  1. .onActivityResult
  2. 首发
  3. upe
  4. thờ
  5. setTitle
  6. ainen
  7. atasets
  8. ginger
  9. -relative
  10. .setPrototypeOf