Neuron Explorer

Qwen3-0.6B

'way' before a how-to phrase

A neuron in layer 11 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

This neuron fires on the word 'way' (or a similar noun or verb) right before a phrase that explains how something is done or why, as in 'an easier way to raise' or 'a formal way of thinking'. It is weakly negative on words like 'pain'. Turning it up made the model repeat simple phrases, which only partly fits.

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

    unique sounds for a reason. There are many stories behind the letters we use so often. The earliest forms of writing evolved because members of ancient civilizations needed more efficient ways to express themselves. Listen to this story to hear about the origins of individual letters

  2. #2 peak

    attention of children. Now the television is in competition with other devices -- computers, video games and smartphones. All of these screens offer entertainment, information, and a way to keep in touch with and stay connected to others.⏎With the arrival of the

  3. #3 peak

    nails. The skin serves many functions, including protecting us from microbes and the elements, helping to regulate body temperature, and receiving sensory

  4. #4 peak

    give us something to look forward to, provide a social outlet, allay worries about health and fitness, increase energy levels, and improve self

  5. #5 peak

    including the railroads, likely contributed to the concentration of these companies within this area. Another factor was a receptive and growing market of customers in need of an easier way to raise

  6. #6 peak

    . The spread of memes has significantly shaped how people communicate, and they have played a fundamental role in advancing the contemporary digital culture. Memes provide a compelling new way of blending a few things such as art, creativity, humor and the intended message

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 key point is that the same point is in the same line (or on the same plane, e.g., on the same line as the point, the line is parallel to the point, and the point is either on the same line

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

    colon due to the intense pressure that comes from a lack of fiber. Diverticulitis is when the pouches become inflamed or infected and cause abdominal pain.⏎Coffee has one of the higher caffeine counts, ranging in caffeine from 6

  2. #2 peak

    is a common sign of appendicitis, especially in children. When the appendix becomes inflamed, it can swell and fill with pus. This can cause severe pain, nausea, and vomiting. Swelling is therefore an important symptom to watch for

  3. #3 peak

    help at Expertsmind⏎Are you searching Operations Management expert for help with Characteristics of Linear Programming questions? Characteristics of Linear Programming topic is not easier to learn without external help? We at

  4. #4 peak

    fight the infection. The increase in white blood cells can lead to a fever. A fever may also be accompanied by other symptoms of appendicitis, such as pain, swelling, and loss of appetite.⏎Gas is another common symptom of appendicit

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. L8:2544 suppresses
  2. L7:1198 suppresses
  3. L8:796 excites
  4. L10:451 excites
  5. L9:1716 weakly excites
  6. L7:1743 weakly suppresses
  7. L3:2789 weakly excites
  8. L10:2920 weakly excites
  9. L10:1219 weakly suppresses
  10. L8:840 weakly suppresses
  11. L5:2141 weakly excites
  12. L8:1184 weakly suppresses
  13. L9:1281 weakly excites
  14. L9:1618 weakly excites
  15. L9:1497 weakly suppresses
  16. L10:2094 weakly excites
  17. L10:2165 weakly excites
  18. L2:2932 weakly suppresses
  19. L10:1308 weakly suppresses
  20. L7:271 weakly excites
  21. L10:2447 weakly excites
  22. L8:1936 weakly excites
  23. L8:1464 weakly suppresses
  24. L10:1856 weakly suppresses

Feeds later layers

  1. L13:678 excites
  2. L12:913 suppresses
  3. L14:2778 excites
  4. L13:2607 excites
  5. L12:1670 excites
  6. L12:493 excites
  7. L13:2577 excites
  8. L13:1363 excites
  9. L12:2283 excites
  10. L12:363 excites
  11. L14:2357 excites
  12. L12:2369 excites
  13. L12:2060 excites
  14. L14:2307 suppresses
  15. L15:3047 excites preposition introducing a definition or date
  16. L14:284 excites
  17. L13:1203 suppresses
  18. L20:2197 weakly excites
  19. L17:1705 weakly excites
  20. L13:2424 weakly suppresses
  21. L15:772 weakly suppresses
  22. L14:2436 weakly excites
  23. L12:288 weakly suppresses
  24. L13:1826 weakly 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 -1.89 and -1.742 values between -1.59 and -1.441 values between -1.44 and -1.292 values between -1.29 and -1.147 values between -1.14 and -0.9913 values between -0.99 and -0.8516 values between -0.85 and -0.7041 values between -0.70 and -0.5584 values between -0.55 and -0.40171 values between -0.40 and -0.25488 values between -0.25 and -0.101915 values between -0.10 and 0.04701 values between 0.04 and 0.19224 values between 0.19 and 0.3485 values between 0.34 and 0.4948 values between 0.49 and 0.6421 values between 0.64 and 0.7919 values between 0.79 and 0.9423 values between 0.94 and 1.0810 values between 1.08 and 1.238 values between 1.23 and 1.386 values between 1.38 and 1.535 values between 1.53 and 1.682 values between 1.68 and 1.832 values between 1.83 and 1.975 values between 1.97 and 2.122 values between 2.12 and 2.271 values between 2.27 and 2.421 values between 2.57 and 2.723 values between 2.86 and 3.011 values between 3.90 and 4.05offnegativepositiveoften

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. 是我们
  2. 녕
  3. zon
  4. 帮助企业
  5. anel
  6. restaur
  7. ocl
  8. alez
  9. 但现在
  10. selves

Suppresses

  1. Petersburg
  2. (le
  3. 游艇
  4. (head
  5. 不同的
  6. ACHER
  7. le
  8. dr
  9. Corinthians
  10. leve