Neuron Explorer

Qwen3-0.6B

'and' between two verbs

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 reacts to the word 'and' when it joins two actions, like 'download and print' or 'identify and alter'. It is a weak signal. Turning it up gave rambling, repetitive text that did not show a clear theme.

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

    your assignment to the M5: Assignment 1 Dropbox.⏎Grading Criteria and Rubric⏎Note: This assignment will be graded using a rubric. Download and read the rubric to understand the expectations.<|endoftext|>A new bionic hand has

  2. #2 peak

    are referring to.⏎For ease of use the language guide is available in full as both a pdf via the link above, and as text below. You may download and print the language guide for personal use.⏎Please note: this is not a static

  3. #3 peak

    in a legible format, making it easier to perform the following actions:⏎- Review changes while staging.⏎- Decide whether to rebase or merge.⏎- Copy and paste code from a file.⏎Giving you even more confidence in your code game.

  4. #4 peak

    package can be installed like so:⏎cabal install bool-extras⏎If you don’t have cabal-install, you can manually download and install bool-extras:⏎tar xzf bool-extras-0.3.

  5. #5 peak

    items | World psychology |⏎Mindfulness-Based Cognitive Therapy (MBCT) is a method of therapy which blends features of two disciplines:⏎- Cognitive therapy aims to identify and alter cognitive distortions (warped or inaccurate thoughts);⏎- Mindfulness is a

  6. #6 peak

    the percentage of a number, you should first know how many decimal places it falls under. You can do this by multiplying the number by the maximum possible mark. Once you’ve determined the maximum, you can also divide it by the total marks.

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: How to best to write an effective, but how to best to write an effective, how to best to write an effective, how to best to write an effective, how to best to write an effective, how to best to write an effective

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

    *A variety of GO Math resources are available.⏎*It can be accessed from school and home.⏎Research shows that students must have fact fluency in order to be successful in higher level math.⏎FASTT - Fluency and Automaticity through

  2. #2 peak

    methanol, or (2) transform biomass into long-chain molecules that can also be used as fuel.⏎CO2 is kinetically and thermodynamically inert, making reduction by chemical hydrogenation

  3. #3 peak

    open up the physical structure of the plant cell walls to permit further deconstruction during the hydrolysis stepyea. The more open structure of the resulting material makes the remaining carbohydrate polymers more accessible for hydrolytic conversion to soluble sugars by

  4. #4 peak

    in a lot of situations because if the physical wound isn't there it could still leave a mark on the mind," Siddens said.⏎The students said

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. L12:1358 suppresses
  2. L8:984 suppresses
  3. L6:2695 suppresses
  4. L13:2493 suppresses
  5. L6:2530 suppresses
  6. L6:2820 weakly suppresses
  7. L8:410 weakly suppresses
  8. L12:1701 weakly excites
  9. L12:2208 weakly excites
  10. L6:2050 weakly suppresses
  11. L6:1500 weakly excites
  12. L14:2944 weakly suppresses
  13. L14:457 weakly suppresses
  14. L6:1185 weakly suppresses
  15. L11:1370 weakly excites 'then' in step-by-step sequences
  16. L11:654 weakly suppresses verb after modal can or could
  17. L12:1522 weakly suppresses comma after a 'before...' clause
  18. L14:2447 weakly excites
  19. L12:1084 weakly suppresses
  20. L9:2121 weakly suppresses
  21. L10:1664 weakly suppresses
  22. L1:1524 weakly excites
  23. L12:2375 weakly suppresses
  24. L12:2888 weakly suppresses

Feeds later layers

  1. L18:2274 excites
  2. L20:1560 suppresses
  3. L19:2236 suppresses
  4. L19:1396 suppresses
  5. L17:1642 excites
  6. L19:3020 suppresses
  7. L17:1698 weakly excites
  8. L16:1593 weakly suppresses
  9. L21:1045 weakly suppresses
  10. L19:517 weakly suppresses
  11. L16:1723 weakly suppresses
  12. L16:2949 weakly suppresses
  13. L27:777 weakly suppresses
  14. L21:250 weakly suppresses
  15. L18:2515 weakly suppresses
  16. L17:691 weakly suppresses
  17. L18:2932 weakly suppresses
  18. L16:1875 weakly suppresses
  19. L19:1349 weakly suppresses
  20. L17:2495 weakly suppresses
  21. L17:413 weakly suppresses
  22. L23:419 weakly suppresses
  23. L18:3056 weakly suppresses
  24. L24:1269 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 -8.37 and -8.121 values between -4.39 and -4.141 values between -4.14 and -3.892 values between -2.90 and -2.653 values between -2.40 and -2.152 values between -2.15 and -1.901 values between -1.90 and -1.654 values between -1.65 and -1.417 values between -1.41 and -1.1619 values between -1.16 and -0.9126 values between -0.91 and -0.6656 values between -0.66 and -0.41302 values between -0.41 and -0.162612 values between -0.16 and 0.09751 values between 0.09 and 0.3485 values between 0.34 and 0.5824 values between 0.58 and 0.834 values between 0.83 and 1.085 values between 1.08 and 1.332 values between 1.33 and 1.58offnegativepositiveoften

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. 除外
  2. bordered
  3. deactivated
  4. ogeneity
  5. finally
  6. (QtCore
  7. requency
  8. jewels
  9. undefeated
  10. IFT

Suppresses

  1. 很难
  2. 难
  3. enan
  4. plx
  5. 难以
  6. asso
  7. compost
  8. comp
  9. -make
  10. ctors