Neuron Explorer

Qwen3-0.6B

'something' as an object

A neuron in layer 17 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 'something' when it follows a verb or preposition, as in 'try something new' or 'the start of something'. Turning it up gave output that did not clearly show this word pattern.

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

    . He also refused to touch the stock of tobacco companies, even when they were all the rage in the 00s. Again, he felt there was something wrong with profiting from a company that essentially poisons its customers. As a

  2. #2 peak

    abbiness surrounding her. Her first social evening is also a disappointment, for she finds the conversation of both men and women personal and trivial. She tries to introduce something different. On the way home, however, she is lectured by her husband

  3. #3 peak

    on their chemistry, and the unique way that individual reactions coalesce into something different in each class section in any given semester. It’s a fluid, organic

  4. #4 peak

    interests, and ultimately motivate your student to tackle summer work with a renewed sense of creativity.⏎Whether students are doing artwork, exploring the world around them, or performing something they made,

  5. #5 peak

    be more ambitious about adding to your activities.⏎Go here for more info⏎There are many ways to fit exercise and physical activity into your regular routine:⏎- Try something new: If you baby-sit for your grandchildren, how about walking to the

  6. #6 peak

    the best defenses - virus protection programs, firewalls and such - but it is almost inevitable. Just as human germs vary, your computer may become infected with something minor - like a trojan - or major - like a worm. When something

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 have to (for the experience) to be in a different country to experience the difference between the other country to know in peace, but also to know the other country to understand, to know it to love, and to know it to

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

    , college, and post-college.⏎To view this DRM protected ebook on your desktop or laptop you will need to have Adobe Digital Editions installed. It is a free software. We also strongly recommend that you sign up for an AdobeID at

  2. #2 peak

    mathematical problems as mathematical problems.⏎To view this DRM protected ebook on your desktop or laptop you will need to have Adobe Digital Editions installed. It is a free software. We also strongly recommend that you sign up for an AdobeID at

  3. #3 peak

    ebook on your desktop or laptop you will need to have Adobe Digital Editions installed. It is a free software. We also strongly recommend that you sign up for an AdobeID at the Adobe website. For more details please see FAQ 1&

  4. #4 peak

    Compute Module actuates a series of motors and grippers, while also running a two-phase algorithm from Herbert Kociemba. Aside from that, the ATmega328 board was employed to control an LDR-based scanner which consists of

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. L11:416 strongly suppresses
  2. L14:1422 strongly suppresses
  3. L6:2443 suppresses
  4. L13:504 suppresses
  5. L16:2715 suppresses
  6. L12:1006 excites
  7. L6:2785 excites
  8. L9:1633 excites
  9. L15:65 excites
  10. L13:1057 excites 'more' after 'a' or 'prefer a'
  11. L9:758 suppresses
  12. L6:479 suppresses
  13. L3:31 excites
  14. L10:39 suppresses
  15. L5:2841 suppresses
  16. L14:1265 suppresses
  17. L8:1801 suppresses
  18. L10:382 excites
  19. L10:363 excites
  20. L3:1090 suppresses
  21. L1:2420 suppresses
  22. L1:1797 suppresses
  23. L8:15 suppresses unclear: mid-phrase words across many languages
  24. L14:1248 weakly suppresses

Feeds later layers

  1. L19:3017 strongly excites
  2. L19:914 strongly suppresses
  3. L20:1991 excites
  4. L22:1798 excites
  5. L19:2331 excites
  6. L20:983 excites
  7. L26:375 excites
  8. L24:704 excites
  9. L19:1620 excites
  10. L20:880 excites
  11. L21:736 excites
  12. L19:1268 excites
  13. L21:225 suppresses
  14. L19:3034 excites
  15. L21:916 excites
  16. L23:473 excites
  17. L19:2208 excites
  18. L24:150 excites
  19. L24:546 excites
  20. L19:1657 excites
  21. L21:1722 excites
  22. L19:249 suppresses
  23. L25:1208 excites
  24. L20:161 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 -3.25 and -2.771 values between -2.77 and -2.293 values between -2.29 and -1.8138 values between -1.81 and -1.33277 values between -1.33 and -0.85814 values between -0.85 and -0.362348 values between -0.36 and 0.12363 values between 0.12 and 0.6032 values between 0.60 and 1.0810 values between 1.08 and 1.566 values between 1.56 and 2.045 values between 2.04 and 2.521 values between 2.52 and 3.012 values between 3.01 and 3.491 values between 3.49 and 3.971 values between 4.93 and 5.411 values between 5.41 and 5.891 values between 5.89 and 6.371 values between 8.78 and 9.261 values between 15.52 and 16.00offnegativepositiveoften

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. ort
  2. transient
  3. non
  4. def
  5. ORT
  6. trig
  7. kin
  8. ir
  9. abstract
  10. prim

Suppresses

  1. Presidency
  2. Equality
  3. Equality
  4. Cities
  5. equality
  6. ivities
  7. 管理水平
  8. ignment
  9. uhan
  10. 껏