Neuron Explorer

Qwen3-0.6B

'than' in comparisons

A neuron in layer 14 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 'than' when a sentence compares two things, such as 'far faster than' or 'more quickly than'. It is quiet, and slightly negative, on words that point to a different one of something, like 'another' or 'some runners'. Turning it down pushed the writing toward the word 'different' again and again.

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

    when we talk about instincts, and isn’t part of the more intellectual decision making processes, indeed it shunts them out of the way, and reacts far faster than we can consciously think.⏎Why does the amygdala want to make a copy

  2. #2 peak

    , had it not been for a tendency toward wild-eyed anti-communism on the American side, the Soviet Union may have collapsed under its own weight much sooner than it did. The stridence and belligerency emanating from Washington,

  3. #3 peak

    :⏎• Increased carbon emissions: As a major contributor to climate change, the burning of oil for heating and cooking purposes releases far higher levels of carbon into the atmosphere than more efficient off-grid fuels like LPG.⏎• Cost: Through a combination of

  4. #4 peak

    India has a trade deficit with China of nearly $50 billion, its largest with any country. Singapore, with a population about 240 times smaller than India, sells twice as many goods to China each year.⏎"Both sides are

  5. #5 peak

    AFs views drawing as a practice of “making” more than “taking” (Taussig 2011). Drawings are far less intrusive than photography, and encourage the necessity to take time, observe, wait, and get

  6. #6 peak

    others increase their happiness, it also increases our own. Studies amongst different cultures and ages have consistently found that individuals who spend money on others report higher levels of happiness than those who spend money on themselves. Researchers suggest that the reward we experience from helping

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 question and the meaning and the meaning of the same and the meaning of one may be different and the meaning of one may be different and the meaning of one may be different and the meaning of one may be different and the meaning of one

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

    0% per week. Keep in mind, though, that everyone is different and that even 10% may be too much of a weekly mileage increase for some runners.⏎The causes of low bone density include genetics, poor dietary calcium or vitamin

  2. #2 peak

    helping your child address ADHD behavioral challenges. Again, each child is different, so it’s important to recognize that what is appropriate and works for one child might not for another. But these are good natural strategies to implement and see if they can help

  3. #3 peak

    or weekly bowel movements. Three bowel movements a day might be normal for one person, while three a week may be normal for another. Also, some people naturally have firmer stools than others.⏎At one time

  4. #4 peak

    concept of learning styles. Some of you wanted to know more about that.⏎Some folks learn a new concept in A&P best by hearing their instructor explain it. Some favor learning it from the textbook. Others need to play with specimens in the lab

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. L12:1912 suppresses
  2. L8:1830 suppresses
  3. L12:2918 excites
  4. L11:662 suppresses
  5. L12:276 suppresses
  6. L7:2400 excites
  7. L13:1768 excites
  8. L11:3023 suppresses gender and race in lists of identities
  9. L11:2186 excites
  10. L13:264 suppresses
  11. L13:818 suppresses
  12. L10:736 weakly excites
  13. L12:3007 weakly suppresses
  14. L12:626 weakly excites
  15. L12:1253 weakly excites
  16. L11:983 weakly suppresses
  17. L2:2776 weakly excites
  18. L7:2733 weakly excites
  19. L12:194 weakly excites
  20. L8:867 weakly suppresses
  21. L6:1251 weakly suppresses
  22. L1:2166 weakly suppresses
  23. L9:2603 weakly suppresses
  24. L9:2407 weakly excites

Feeds later layers

  1. L15:1922 suppresses
  2. L16:2807 suppresses
  3. L26:1060 suppresses
  4. L21:2973 suppresses
  5. L15:2440 suppresses word before 'and' or 'or' in pairs
  6. L23:875 suppresses
  7. L19:2976 suppresses
  8. L17:2541 suppresses comma before but also in not only constructions
  9. L15:2448 suppresses
  10. L25:648 suppresses
  11. L16:2051 suppresses
  12. L15:2002 suppresses
  13. L19:2205 suppresses
  14. L21:1090 suppresses
  15. L18:2942 suppresses
  16. L18:2761 suppresses
  17. L19:2308 suppresses
  18. L21:2296 suppresses
  19. L15:2015 suppresses
  20. L21:1056 suppresses
  21. L15:1585 suppresses
  22. L15:454 excites
  23. L17:1036 suppresses
  24. L20:944 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 -1.85 and -1.772 values between -1.68 and -1.601 values between -1.52 and -1.433 values between -1.35 and -1.261 values between -1.18 and -1.092 values between -1.01 and -0.925 values between -0.92 and -0.844 values between -0.84 and -0.761 values between -0.76 and -0.673 values between -0.67 and -0.599 values between -0.59 and -0.5025 values between -0.50 and -0.4266 values between -0.42 and -0.33193 values between -0.33 and -0.25413 values between -0.25 and -0.17642 values between -0.17 and -0.081119 values between -0.08 and 0.00854 values between 0.00 and 0.09312 values between 0.09 and 0.17158 values between 0.17 and 0.2654 values between 0.26 and 0.3411 values between 0.34 and 0.438 values between 0.43 and 0.514 values between 0.51 and 0.594 values between 0.59 and 0.683 values between 0.68 and 0.764 values between 0.76 and 0.851 values between 0.85 and 0.931 values between 1.02 and 1.101 values between 1.27 and 1.352 values between 1.44 and 1.52offnegativepositiveoften

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. built
  2. iom
  3. $__
  4. noun
  5. built
  6. pl
  7. ��
  8. acknowledge
  9. وم
  10. :\/\/

Suppresses

  1. 不同的
  2. 有不同的
  3. 不同
  4. 有所不同
  5. _different
  6. 不一样
  7. 不一样的
  8. Different
  9. different
  10. different