Neuron Explorer

Qwen3-0.6B

"the" after verbs like laid, reaching, raises

A neuron in layer 20 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" right after verbs such as "laid", "reaching" or "raises", where a noun like "foundation" or "possibility" is expected next. Turned up, the model drifted into unrelated topics and did not clearly echo this pattern.

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

    programs (Katki et al., 2011) have informed these revised guidelines. More information on HPV and cervical cancer.⏎DCEG research provided the initial assessment of the specificity, sensitivity, and appropriate applications of the first-generation HIV

  2. #2 peak

    , and participation—cover the necessities to produce generations of environmental stewards. These objectives, coupled with the declaration’s focus on creating sustainably built environments, laid the foundation for effective environmental education that is still utilized today.⏎As environmental educators, we

  3. #3 peak

    foreign tongues. This “Pentecostal” experience allowed the apostles direct communication with God, which signaled a major shift in the religious landscape and laid the foundation for what would become Christianity. You’ll notice that the disciples are always depicted

  4. #4 peak

    established HPV as the necessary cause of this malignancy (Schiffman et al., 1993, 2007). Their efforts laid the groundwork for vaccine development and improved strategies for screening.⏎The licensure of prophyl

  5. #5 peak

    as RF, this range extends to 300 GHz.⏎RF and wireless have been around for over a century with Alexander Popov and Sir Oliver Lodge laying the groundwork for Guglielmo Marconi's wireless radio developments in the early

  6. #6 peak

    , no longer holds true. Some extremely complex and costly engineering tricks promise to help shrink the size of components etched into silicon chips further, but engineers are reaching the physical limits of how small these components, which have features measured in billionths of

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: there is a new interest in the work from the people from the past. That from what I see, there is there is much in that from the people, not from the wrong, and there is so much in there. So there is

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

    communication between classmates before the school year starts.⏎Discover the World Through Virtual Tours⏎Thanks to technological advancements, students from kindergarten to high school can explore the world from the comfort of their own homes. Take advantage of virtual tours offered by museums, historical

  2. #2 peak

    <|endoftext|>We live in an increasingly online world. You can trade stocks, buy groceries, pay bills - almost any financial transaction you need to make can be done in the comfort of your own home, at your computer. With identity theft posing a real

  3. #3 peak

    2015, 54 of Atlanta Public Schools’ 81 schools implemented CEP<|endoftext|>I’ve been a longtime fan of Understanding Science from the University of California Museum of Paleontology at Berkeley. It’s a comprehensive resource for

  4. #4 peak

    buttons, alligator clip pads and more. Build projects with Circuit Playground in a few minutes with the drag-and-drop MakeCode programming site, learn computer science using the CS Discoveries class on code.org, jump into CircuitPython to learn Python and

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. L17:30 suppresses
  2. L9:24 excites "the" before a named organisation or title
  3. L18:799 excites
  4. L10:120 suppresses
  5. L10:1377 suppresses
  6. L9:662 suppresses
  7. L19:2283 suppresses
  8. L16:792 excites "the" inside a prepositional phrase
  9. L16:419 excites
  10. L18:171 excites
  11. L11:1168 excites
  12. L4:52 excites
  13. L4:73 suppresses
  14. L6:61 suppresses
  15. L2:205 excites
  16. L15:2120 suppresses
  17. L11:331 excites
  18. L13:1912 weakly excites
  19. L9:2198 weakly excites
  20. L10:1644 weakly excites
  21. L3:291 weakly excites
  22. L19:427 weakly suppresses
  23. L19:1023 weakly excites
  24. L16:1594 weakly excites

Feeds later layers

  1. L24:2497 excites
  2. L26:720 weakly suppresses 'New' in place names
  3. L21:172 weakly suppresses
  4. L23:439 weakly suppresses
  5. L25:381 weakly suppresses
  6. L21:1040 weakly suppresses
  7. L26:1373 weakly excites
  8. L26:2648 weakly excites
  9. L26:989 weakly suppresses
  10. L25:1059 weakly excites
  11. L27:596 weakly suppresses
  12. L24:590 weakly suppresses
  13. L25:1206 weakly suppresses
  14. L22:1664 weakly excites
  15. L24:936 weakly suppresses
  16. L27:2396 weakly excites
  17. L27:814 weakly excites
  18. L26:897 weakly suppresses the 'gro' fragment in words
  19. L21:816 weakly excites
  20. L26:115 weakly suppresses
  21. L22:878 weakly suppresses
  22. L21:252 weakly excites
  23. L23:143 weakly excites
  24. L23:785 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 -18.84 and -17.761 values between -17.76 and -16.681 values between -16.68 and -15.601 values between -15.60 and -14.521 values between -14.52 and -13.442 values between -13.44 and -12.362 values between -12.36 and -11.282 values between -11.28 and -10.203 values between -10.20 and -9.123 values between -9.12 and -8.046 values between -8.04 and -6.967 values between -6.96 and -5.888 values between -5.88 and -4.8012 values between -4.80 and -3.7212 values between -3.72 and -2.6418 values between -2.64 and -1.56265 values between -1.56 and -0.483334 values between -0.48 and 0.6196 values between 0.61 and 1.6937 values between 1.69 and 2.7721 values between 2.77 and 3.8514 values between 3.85 and 4.9315 values between 4.93 and 6.018 values between 6.01 and 7.0912 values between 7.09 and 8.177 values between 8.17 and 9.256 values between 9.25 and 10.333 values between 10.33 and 11.412 values between 11.41 and 12.494 values between 13.57 and 14.651 values between 14.65 and 15.731 values between 16.81 and 17.891 values between 18.97 and 20.051 values between 23.30 and 24.38offnegativepositiveoften

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. nature
  2. ";}⏎
  3. 的
  4. nature
  5. ";}
  6. 的前提
  7. jh
  8. ');?>⏎
  9. portion
  10. edar

Suppresses

  1. 他们的
  2. their
  3. Their
  4. insights
  5. leurs
  6. 坏事
  7. 他们在
  8. ấy
  9. !",
  10. Their