Neuron Explorer

Qwen3-0.6B

"of" after "kind" or "sort"

A neuron in layer 1 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 "of" in phrases like "some kind of" or "some sort of". Those phrases are the spot where the writer is about to name the type of something. Turning it up did not give a clear topic change in the samples.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    , after Hawaii became a state the prior year."⏎From the Internet<|endoftext|>Fraud is a term that is used to describe deception perpetrated in order to achieve some kind of goal or benefit. Fraud can be either civil, criminal, or both, with

  2. #2 peak

    natural disaster in an attempt to help the people in the midst of crisis. Individual aid groups have also developed treatment programs for women and men who have suffered some kind of sexual abuse. However, mental healthcare in developing countries has never been taken on by

  3. #3 peak

    .⏎For example, let’s say a study tried to test the effects of violent game by having a participant play a game and then asking them to do some kind of word completion test. If the study found that people who played the violent game were

  4. #4 peak

    royalty (for other<|endoftext|>There is often an ordinary, unremarkable beginning to a life dependent on opioids. Someone might experience a routine surgery or have some kind of accident such as a fall or car crash. Following events like these, a

  5. #5 peak

    a few days, the CNT and the Socialist General Confederation of Workers (UGT), which since December 1933 had been discussing some kind of working-class alliance, planned a national general strike. It began on October 4

  6. #6 peak

    and also points to separate research which estimates that up to 22% of crashes could be caused by some kind of distraction; and drivers who perform a secondary task at the wheel, like using a

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: I have been thinking about recently, and I have been thinking about recently. I have been thinking about recently. I have been thinking about recently. I have been thinking about recently. I have been thinking about recently. I have been thinking about

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

    a goodly portion of those who wished to live outside the city during their visit. It would

  2. #2 peak

    In addition to a few reminders of the Middle Ages, the upper town is best-known for its 16th-century Renaissance heritage. This was the most opulent of all periods in the town’s history from the political and cultural point of view

  3. #3 peak

    Electronic Privacy Information Center, 1992.⏎The American Inquisition: Federalism Gone Amok⏎Modern inquisitorial methodology has changed quite a lot since the time of the Spanish Inquisition<|endoftext|>Fire can incite awe, terror

  4. #4 peak

    or metal in an attempt to straighten their back.⏎Dr Lund said: "Scoliosis is a painful illness, and Richard would have been in quite a lot of pain on a daily basis.⏎"These methods could also

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. L0:180 weakly excites
  2. L0:645 weakly excites
  3. L0:405 weakly suppresses
  4. L0:458 weakly excites
  5. L0:1028 weakly excites
  6. L0:1732 weakly suppresses
  7. L0:1805 weakly suppresses
  8. L0:1332 weakly suppresses
  9. L0:1730 weakly suppresses
  10. L0:437 weakly suppresses
  11. L0:2792 weakly suppresses middle pieces of split words
  12. L0:1228 weakly suppresses
  13. L0:753 weakly excites
  14. L0:1279 weakly excites
  15. L0:289 weakly suppresses
  16. L0:84 weakly suppresses
  17. L0:97 weakly excites
  18. L0:2868 weakly excites
  19. L0:2038 weakly suppresses
  20. L0:2394 weakly suppresses
  21. L0:2517 weakly suppresses
  22. L0:620 weakly excites
  23. L0:287 weakly suppresses
  24. L0:305 weakly excites

Feeds later layers

  1. L20:2575 weakly suppresses
  2. L14:2859 weakly excites
  3. L6:2475 weakly suppresses
  4. L17:1540 weakly suppresses
  5. L14:2917 weakly excites
  6. L23:747 weakly suppresses
  7. L7:116 weakly suppresses
  8. L20:2842 weakly excites
  9. L21:1344 weakly suppresses
  10. L11:2623 weakly suppresses
  11. L13:22 weakly suppresses
  12. L18:1709 weakly excites
  13. L4:2061 weakly excites
  14. L25:1097 weakly excites 'the' or 'of' after a role or title
  15. L14:1438 weakly suppresses
  16. L13:3000 weakly suppresses
  17. L23:2150 weakly suppresses
  18. L4:1873 weakly excites
  19. L2:87 weakly suppresses
  20. L5:3009 weakly suppresses 'to' after a verb or noun
  21. L11:2937 weakly excites
  22. L11:1078 weakly suppresses
  23. L20:1156 weakly suppresses
  24. L3:1550 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 -1.79 and -1.652 values between -1.10 and -0.965 values between -0.96 and -0.8210 values between -0.82 and -0.6816 values between -0.68 and -0.5431 values between -0.54 and -0.4072 values between -0.40 and -0.26291 values between -0.26 and -0.121923 values between -0.12 and 0.021332 values between 0.02 and 0.16163 values between 0.16 and 0.3020 values between 0.30 and 0.4410 values between 0.44 and 0.585 values between 0.58 and 0.727 values between 0.72 and 0.863 values between 0.86 and 1.001 values between 1.00 and 1.142 values between 1.14 and 1.281 values between 1.28 and 1.422 values between 1.42 and 1.564 values between 1.56 and 1.703 values between 1.70 and 1.831 values between 1.97 and 2.111 values between 2.25 and 2.391 values between 2.67 and 2.811 values between 3.65 and 3.79offnegativepositiveoften

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. vant
  2. 률
  3. rement
  4. 蛙
  5. あります
  6. 问道
  7. |r
  8. Wonderland
  9. na
  10. ROUND

Suppresses

  1. 同样
  2. CharCode
  3. bury
  4. .setFill
  5. itos
  6. 了一口
  7. LoginPage
  8. arsen
  9. 当事
  10. ős