Neuron Explorer

Qwen3-0.6B

superlatives like 'largest' and 'first'

A neuron in layer 16 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 lights up on words that single out something as the top example, like 'largest', 'biggest', 'first' and 'most', especially after phrases like 'one of the' or 'the very'. Turning it up made the model drift into unrelated, repetitive text, so the clearest evidence is in the examples themselves.

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

    all men. Prostate cancer remains an important public health concern in Western countries and an emerging malignancy in developing nations, and preceded only by Lung cancer, Prostate cancer is the leading cause of cancer death among men. According to the American Cancer

  2. #2 peak

    This is closely followed by the premolars appearing around the 3rd to 4th week. It could take a few weeks longer for some cats, but most kittens will have their baby teeth out by the 8th week.⏎Check;

  3. #3 peak

    responsible for 17 percent of the protein we consume<|endoftext|>The first non-stop trans-Atlantic flight⏎THE VERY FIRST NON-STOP TRANS-ATLANTIC FLIGHT⏎Note: The U

  4. #4 peak

    times the chance of developing PD. This increased risk is most likely because of a combination of environmental and genetic factors.⏎Environmental causes are being researched and the strong consistent findings are that rural living, exposure to well water, and exposure

  5. #5 peak

    250 tons of steel and 4,000 panes of glass. The dome spans over 200 feet and is one of the largest free-spanning domes in the world.⏎Despite its grandeur, the hotel

  6. #6 peak

    July 10 and July 12, a massive iceberg broke off from Antarctica’s Larsen C Ice Shelf and it is considered to be one of the biggest icebergs that have ever been observed. The area of the iceberg spreads over

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 need for a new approach to the use of these items. All these items are all just as they are. There are all these all these all. and all these all. and all these all. and all these

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

    the evidence to fit with the preconceived views and aspirations of their supporters, as well as their own existing political mantras. To some, evidence becomes nothing more than any “fact” (whether true or not) that can be used to

  2. #2 peak

    Large Scale Biosphere-Atmosphere|⏎|Time Period||1985-01-01 to 2001-12-31|⏎This data set includes

  3. #3 peak

    and the glory of the Lord shone around them, and they were filled with great fear. 10 And the angel said to them, “Fear not, for behold, I bring you good news of

  4. #4 peak

    Some philosophers who reject this metaphysical view have swung in the other direction and rejected the idea of the self altogether. David Hume said that the self is nothing more than a bundle of perceptions, and Daniel Dennett dismissed the self as merely a

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. L14:961 strongly suppresses
  2. L14:816 strongly suppresses
  3. L14:858 suppresses
  4. L15:945 suppresses
  5. L15:308 suppresses
  6. L13:452 suppresses
  7. L15:2845 suppresses
  8. L13:579 suppresses
  9. L13:2806 suppresses
  10. L12:2843 suppresses
  11. L15:1032 suppresses
  12. L12:1753 excites
  13. L8:47 suppresses
  14. L13:1020 excites
  15. L12:1097 suppresses
  16. L13:970 excites
  17. L14:1258 suppresses
  18. L10:171 excites
  19. L13:372 excites
  20. L14:1130 suppresses
  21. L15:1813 suppresses
  22. L13:1789 suppresses
  23. L14:1633 excites
  24. L13:1627 excites

Feeds later layers

  1. L17:521 strongly excites
  2. L18:532 suppresses
  3. L17:595 suppresses
  4. L17:1152 suppresses
  5. L19:2416 suppresses
  6. L18:1482 suppresses
  7. L17:1569 suppresses
  8. L18:1856 suppresses
  9. L17:459 suppresses
  10. L17:2155 suppresses
  11. L18:1386 weakly suppresses unclear: words before 'and' or 'or'
  12. L19:143 weakly excites
  13. L17:144 weakly suppresses
  14. L19:2493 weakly suppresses
  15. L19:392 weakly excites
  16. L19:1431 weakly suppresses
  17. L17:2569 weakly suppresses
  18. L19:1424 weakly suppresses
  19. L17:1517 weakly suppresses
  20. L17:37 weakly suppresses
  21. L18:1560 weakly excites
  22. L18:619 weakly suppresses
  23. L19:653 weakly suppresses
  24. L19:469 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 -2.63 and -2.453 values between -2.45 and -2.261 values between -2.26 and -2.086 values between -2.08 and -1.893 values between -1.89 and -1.717 values between -1.71 and -1.5212 values between -1.52 and -1.3423 values between -1.34 and -1.1533 values between -1.15 and -0.9757 values between -0.97 and -0.7993 values between -0.79 and -0.60168 values between -0.60 and -0.42232 values between -0.42 and -0.23310 values between -0.23 and -0.05438 values between -0.05 and 0.14503 values between 0.14 and 0.32465 values between 0.32 and 0.51395 values between 0.51 and 0.69314 values between 0.69 and 0.88245 values between 0.88 and 1.06220 values between 1.06 and 1.25116 values between 1.25 and 1.4397 values between 1.43 and 1.6252 values between 1.62 and 1.8035 values between 1.80 and 1.9930 values between 1.99 and 2.1720 values between 2.17 and 2.3513 values between 2.35 and 2.547 values between 2.54 and 2.724 values between 2.72 and 2.913 values between 2.91 and 3.091 values between 3.65 and 3.831 values between 4.57 and 4.75offnegativepositiveoften

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. sov
  2. peare
  3. trừ
  4. ば
  5. **
  6. *[
  7. 玩家来说
  8. *\
  9. でしょうか
  10. *t

Suppresses

  1. ataire
  2. OTT
  3. 他也
  4. Platt
  5. postfix
  6. slot
  7. Interpolator
  8. ifice
  9. woord
  10. "));⏎⏎