Neuron Explorer

Qwen3-0.6B

word before 'needed' or 'required'

A neuron in layer 27 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 just before a spot where 'needed' or 'required' would fit, like 'no special preparation is' or 'they simply'. Turning it down made the model repeat the word 'required' over and over.

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

    and well-to-do could afford to own slaves, and they had no reason or need to justify slavery based on any notion of race or racial inferiority. They simply needed slave labor to maintain their “luxurious way of life.”⏎Africans

  2. #2 peak

    of the rich. Their lives were dependent primarily on obedience to their employers. Willingly or not, they participated in a continuing economic system that benefitted primarily those who were super-rich. They were totally bound to those who literally owned them.⏎In

  3. #3 peak

    personnel. Despite looking out of this world, it was a very real aircraft.⏎The D-21, constructed without landing gears, was an innovative contraption that could only be used only once and was jettisoned mid-air from another

  4. #4 peak

    to look for:⏎No special preparation is needed before the test. A chest x-ray will be performed before and after the

  5. #5 peak

    , but more often women would take the avenue that was available to them: writing in letters or editorials in the women’s pages, or under pseudonyms if they managed to break into the main pages of the paper (Rex x). According

  6. #6 peak

    viewed from this PDF file:⏎The HOTBOX uses the principle of insulated cooking. If you can keep the heat that is used to cook food, no replacement heat is necessary to complete

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: required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required required

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

    originate indoors are getting trapped inside.⏎“People normally<|endoftext|>Spring Dawn by Meng Haoran 春眠不觉晓 处处闻啼茑 夜来风雨声 花落知多少⏎Pin Yin Version Chun mian bu jue

  2. #2 peak

    to live in them with ease.⏎There are currently 11 national parks in the Thar Desert, the largest of which are the Rann of Kutch and the Nara Desert Wildlife Sanctuaries. The Rann of K

  3. #3 peak

    aces one still is one of London’s most famous landmarks: the Tower of London.⏎The Tudor period (Henry VIII, Queen Elisabeth I) was one of the most productive in the history of the English monarch

  4. #4 peak

    Tertiary e) limestone at the bottom overlying the basalts. What remains today is a shallow water atoll about 10 kilometers across.⏎Location: As its name suggests, Midway lays nearly half-way between the continents of North America

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. L22:912 strongly excites
  2. L25:1711 strongly suppresses
  3. L21:2956 suppresses
  4. L20:1743 excites
  5. L23:1338 suppresses
  6. L21:1990 excites
  7. L23:108 suppresses
  8. L26:1564 excites
  9. L21:914 excites
  10. L21:1611 excites
  11. L21:1422 suppresses
  12. L19:2621 suppresses
  13. L20:2032 suppresses
  14. L21:1646 excites
  15. L23:127 suppresses
  16. L23:360 suppresses
  17. L26:466 excites
  18. L22:444 suppresses
  19. L18:2747 suppresses
  20. L23:747 excites
  21. L24:422 excites
  22. L19:1087 excites
  23. L24:337 suppresses
  24. L23:975 excites

Feeds later layers

None: this is the last layer.

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 -14.29 and -13.231 values between -12.16 and -11.101 values between -11.10 and -10.031 values between -10.03 and -8.973 values between -8.97 and -7.903 values between -7.90 and -6.845 values between -6.84 and -5.7810 values between -5.78 and -4.7119 values between -4.71 and -3.6547 values between -3.65 and -2.5891 values between -2.58 and -1.52304 values between -1.52 and -0.451836 values between -0.45 and 0.611383 values between 0.61 and 1.6879 values between 1.68 and 2.7426 values between 2.74 and 3.8014 values between 3.80 and 4.8716 values between 4.87 and 5.9313 values between 5.93 and 7.006 values between 7.00 and 8.066 values between 8.06 and 9.134 values between 9.13 and 10.193 values between 10.19 and 11.259 values between 11.25 and 12.325 values between 12.32 and 13.382 values between 13.38 and 14.452 values between 14.45 and 15.511 values between 15.51 and 16.571 values between 17.64 and 18.705 values between 18.70 and 19.771 values between 19.77 and 20.831 values between 21.90 and 22.961 values between 22.96 and 24.021 values between 25.09 and 26.153 values between 26.15 and 27.223 values between 27.22 and 28.28offnegativepositiveoften

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. gie
  2. .inspect
  3. 说完
  4. setSupportActionBar
  5. vation
  6. 埤
  7. 概况
  8. 挫
  9. oxy
  10. bite

Suppresses

  1. required
  2. needed
  3. required
  4. Required
  5. Required
  6. needed
  7. Needed
  8. desired
  9. →required
  10. 所需