Neuron Explorer

Qwen3-0.6B

'under' and 'cr' word fragments

A neuron in layer 26 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 fires on the word 'under' when it is part of a phrase like 'underlie' or 'under 5', and it fires on the 'cr' piece of words like 'crimp' or 'crunched'. Turned down, the model kept repeating the word 'under'.

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

    , protected areas and ecological networks. There is no single correct way to apply the ecosystem approach to management of land, water, and living resources. The principles that underlie the ecosystem

  2. #2 peak

    the amount of random error involved in measuring student performance. The risk is that teachers and schools may be wrongfully rewarded or punished because value-added techniques either over- or underestimated their students’ learning gains. However, no intelligent users of standardized testing would

  3. #3 peak

    recent cognitive acquisitions derive from older representational functions of which we are hardly aware. Consideration of musical art can help bring to light the more ancient cognitive functions that underlie modern human cognition.⏎About the Author⏎Charles O. Nussbaum is

  4. #4 peak

    under selective harvesting at medium intensity, followed by selective harvesting at low intensity. The plant community stability was far from the point of stability (20/80) under selective harvesting at high and extra-high intensities, as well as with clear

  5. #5 peak

    , but comparable significant cooling benefits for crop yields by all four methods. Overall, global yields increase ~10% under the three SGs and decrease 5% under emissions reduction, the latter primarily due to reduced CO2 fertilization, relative

  6. #6 peak

    operations depend on molecular processes such as the operations of neurotransmitters and<|endoftext|>Children under 5 getting excessive screen time, study finds⏎A new study shows that most children under 5 years old are getting more screen time than the recommended amount.⏎The

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: under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under under

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

    als we cut down. We exhale long streamers of air back toward the surface, and move slowly as we pass the shared pliers around, cutting and crimping monofilament with the slow-motion care of astronauts maneuvering in zero

  2. #2 peak

    for a whole host of different nutrients, including beta carotene and vitamin C. The numbers were then crunched to see what nutrients were or were not associated with what kinds of breast cancer

  3. #3 peak

    so you get a mini greenhouse for nothing!⏎- milk cartons – just cut off the gabled top⏎- newspaper rolls – roll up newspaper strips and then crimp the bottom⏎- plant pots – if you do buy some transplants,

  4. #4 peak

    crab family. As spines cover the tops of their bodies, spiny king crabs are also called porcupine crabs. These spines not only distinguish

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. L24:784 strongly suppresses
  2. L25:1888 strongly excites
  3. L25:1381 strongly excites
  4. L25:2160 strongly excites
  5. L22:2633 strongly suppresses
  6. L25:3005 strongly suppresses
  7. L22:2064 strongly excites
  8. L25:2502 strongly suppresses
  9. L25:2631 excites
  10. L22:1701 excites
  11. L21:2180 excites
  12. L23:1354 suppresses
  13. L20:1903 excites
  14. L23:2148 excites
  15. L23:640 excites
  16. L22:2518 suppresses
  17. L22:1201 excites
  18. L21:140 suppresses
  19. L23:1688 suppresses
  20. L19:1545 suppresses
  21. L20:706 suppresses
  22. L24:2746 weakly excites
  23. L16:507 weakly excites
  24. L23:476 weakly suppresses

Feeds later layers

  1. L27:1711 suppresses
  2. L27:1001 suppresses
  3. L27:636 weakly suppresses
  4. L27:1700 weakly suppresses
  5. L27:593 weakly excites
  6. L27:2394 weakly suppresses
  7. L27:29 weakly excites
  8. L27:2005 weakly excites
  9. L27:1127 weakly suppresses
  10. L27:1982 weakly suppresses
  11. L27:405 weakly suppresses
  12. L27:2322 weakly suppresses
  13. L27:2422 weakly excites
  14. L27:599 weakly suppresses
  15. L27:2179 weakly suppresses
  16. L27:1902 weakly suppresses
  17. L27:622 weakly suppresses
  18. L27:205 weakly suppresses
  19. L27:2353 weakly suppresses
  20. L27:2335 weakly suppresses
  21. L27:220 weakly suppresses end-of-text token between documents
  22. L27:2402 weakly excites
  23. L27:1558 weakly excites
  24. L27:1819 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 -22.89 and -21.871 values between -12.67 and -11.652 values between -11.65 and -10.621 values between -10.62 and -9.602 values between -9.60 and -8.588 values between -8.58 and -7.563 values between -7.56 and -6.532 values between -6.53 and -5.5110 values between -5.51 and -4.4912 values between -4.49 and -3.4728 values between -3.47 and -2.4547 values between -2.45 and -1.42187 values between -1.42 and -0.402427 values between -0.40 and 0.621145 values between 0.62 and 1.6420 values between 1.64 and 2.673 values between 2.67 and 3.692 values between 3.69 and 4.711 values between 4.71 and 5.733 values between 5.73 and 6.761 values between 6.76 and 7.781 values between 7.78 and 8.801 values between 16.98 and 18.00offnegativepositiveoften

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. alen
  2. ogle
  3. entic
  4. *----------------------------------------------------------------
  5. :+:
  6. apur
  7. ภาค
  8. enzie
  9. 释放
  10. warm

Suppresses

  1. under
  2. Under
  3. Under
  4. under
  5. UNDER
  6. _under
  7. UNDER
  8. -under
  9. _UNDER
  10. bajo