Neuron Explorer

Qwen3-0.6B

'for' after a span of time

A neuron in layer 19 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

The neuron lights up on the word 'for' when it comes right after a length of time, as in 'for several weeks' or 'for years'. When it was turned up, the model kept drifting into talk about how long something lasts, such as 'for a long time' and 'a few days'.

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

    the city was almost destroyed. The conflict spread to outlying provinces.⏎By 1530, the wars of the Sengoku Period had been fought for over two generations. At that point, there were nine major

  2. #2 peak

    only hinted at by the book.⏎I have admired Dick’s work for years, but this is his best. He delves more thoroughly into many themes

  3. #3 peak

    has helped me to uncover student misconceptions and look into their thought processes regularly. A supervisor gave me the book in August, and it sat on my nightstand for several weeks. In my mind, it was going to be another book about visual

  4. #4 peak

    mental illness, affecting more than 350 million people worldwide, according to a Reuters video. Bob Holmes is one of them. "I struggled with that for many years, didn’t know really what to do, tried to pull myself through

  5. #5 peak

    England, in 1813 and made a career for himself as a renowned architect, engineer and surveyor. After living in Dublin for several years in the early 1830s, Lanyon moved

  6. #6 peak

    and challenges for these individuals truly is a spectrum that can range from a completely nonverbal kindergartner, to a professor of linguistics.⏎We’ve known for more than 20 years that the lion’s share of the risk for developing

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: A person who is not a regular, but a regular in a regular, and in a regular of a regular of a regular... a regular of a regular, and a regular of a regular of a regular... a regular of a regular,

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

    Assessment's Biogeography Branch (BB) has been conducting research in Puerto Rico and the US Virgin Islands since 2000 and 2

  2. #2 peak

    process improvements are elastic and, unless rigorously monitored, drift back to their original state. Determin<|endoftext|>Archeological sites are places where people have lived or worked in the past and

  3. #3 peak

    <|endoftext|>The term “education technology” (or “edtech,” as it’s commonly known) refers to more than just learning via computer. Edtech has existed in various forms for as long as education itself has been around. At the beginning of

  4. #4 peak

    .⏎The Hasinai group of the Caddo tribe built a village in the area about 800 A.D. and continued to live in the area until the 1830s, when they migrated to the

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. L15:1825 strongly excites
  2. L17:1842 strongly suppresses
  3. L14:2236 strongly suppresses
  4. L12:2377 excites
  5. L16:1277 excites
  6. L18:2735 excites
  7. L17:2544 suppresses
  8. L18:2437 suppresses
  9. L18:2130 suppresses
  10. L13:2868 suppresses
  11. L17:3010 excites
  12. L15:1520 excites
  13. L14:1088 excites
  14. L13:2890 excites "still" or "as long as" meaning continuing
  15. L18:2977 weakly suppresses
  16. L17:1041 weakly excites
  17. L10:2361 weakly suppresses
  18. L12:39 weakly suppresses
  19. L15:2665 weakly suppresses
  20. L7:1114 weakly suppresses
  21. L18:1505 weakly suppresses
  22. L16:2173 weakly excites
  23. L6:1343 weakly suppresses
  24. L12:2782 weakly suppresses

Feeds later layers

  1. L21:1284 strongly excites
  2. L23:2386 strongly excites
  3. L24:1778 strongly excites 'a' before a time span
  4. L20:1355 strongly excites
  5. L24:1394 excites
  6. L22:1699 excites
  7. L22:2704 excites
  8. L24:587 excites
  9. L20:3050 excites
  10. L23:2777 excites
  11. L27:1469 excites
  12. L23:1919 excites
  13. L22:1074 excites
  14. L20:2745 excites
  15. L22:2601 excites
  16. L20:1219 excites
  17. L21:785 excites
  18. L21:3058 excites
  19. L20:2376 excites "the" after "over" in time spans
  20. L21:848 excites
  21. L27:1131 excites
  22. L21:1797 excites
  23. L24:826 excites
  24. L21:676 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.

5 values between -3.24 and -2.663 values between -2.66 and -2.0810 values between -2.08 and -1.4967 values between -1.49 and -0.91869 values between -0.91 and -0.332522 values between -0.33 and 0.25359 values between 0.25 and 0.8434 values between 0.84 and 1.4212 values between 1.42 and 2.004 values between 2.00 and 2.585 values between 2.58 and 3.162 values between 3.16 and 3.753 values between 4.33 and 4.913 values between 5.49 and 6.082 values between 6.08 and 6.662 values between 7.24 and 7.821 values between 7.82 and 8.401 values between 8.40 and 8.991 values between 8.99 and 9.571 values between 11.90 and 12.481 values between 16.55 and 17.141 values between 19.47 and 20.05offnegativepositiveoften

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. 一段时间
  2. 很长时间
  3. years
  4. 几年
  5. months
  6. 两年
  7. 多久
  8. 很久
  9. 太久
  10. 三个月

Suppresses

  1. uctose
  2. _language
  3. твержда
  4. ovich
  5. _scaling
  6. ��
  7. юр
  8. .language
  9. ítica
  10. _positive