Neuron Explorer

Qwen3-0.6B

the word 'National' in park names

A neuron in layer 8 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 'National' in place names like 'Grand Staircase-Escalante National Monument' or 'Everglades National Park'. Turning it up did not bring out park talk in the model's writing, so the link is only visible in the text it reads.

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

    fossils from Utah.⏎Below, click on the various districts (West Desert, Green River, Color Country, Canyon Country, or Grand Staircase-Escalante National Monument) to find out more about paleontology in that area.<|endoftext|>The principal

  2. #2 peak

    overlook on the rim of the Grand Canyon?⏎Point Imperial, 8,803 feet⏎How many species of mosquitos inhabit the Everglades National Park?⏎43, but only 12 bite⏎What park has the

  3. #3 peak

    get a new generation interested in using the parks, much less trying a decidedly "low-tech" activity like camping?⏎Rangers at Lake Mead National Recreation Area are trying one approach with high school students in the Las Vegas area,

  4. #4 peak

    ants were perfect and completely green and had a good amount<|endoftext|>A remarkable new species of tyrannosaur has been unearthed in Grand Staircase-Escalante National Monument (GSENM), southern Utah. The huge carnivore inhabited Laram

  5. #5 peak

    hinoceros unicorns),⏎- Pigmy Hog (Sus salvanius).⏎Over 426 bird species are reported to be found in Royal Manas National Park:⏎- Great Hornbill (Buceros bicornis),⏎- Ruf

  6. #6 peak

    the ecosystems they inhabit, protected areas still fall short of protecting species and their ecological needs. In many instances, protected areas such as Nouabalé-Ndoki National Park in The Republic of Congo do not cover the full range 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: I would like to know how to do the following? What is the purpose of this exercise? Okay, so I'm trying to find out how to do the following, and I would like to know how to do the first part of the

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

    them.⏎The horse-back journey to Fort Assiniboine took six days. There the party once more took to canoes which were worked up the Athabasca River to Jasper House. Two days after leaving the fort, the oldest Lebl

  2. #2 peak

    .⏎The Wilkes series consists of shallow, well drained soils with moderately slow permeability. These soils

  3. #3 peak

    itzer Dam.⏎- A network of 150 miles of trails, including the Wild Oak National Recreation Trail and a portion of the Great Eastern Trail. The historic Civilian Conservation Corps (CCC) built many of the trails

  4. #4 peak

    to a place where there is a riffle in the water. Mora is located a mile from the mouth of the Quillayute River at its confluence with the Dickey River or Dickodochteda, meaning muddy water in

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. L6:2455 suppresses
  2. L7:1583 excites
  3. L3:3042 suppresses
  4. L3:1216 suppresses
  5. L5:1805 suppresses
  6. L7:2529 suppresses
  7. L7:373 weakly suppresses
  8. L4:1376 weakly excites
  9. L6:1886 weakly suppresses
  10. L6:1742 weakly suppresses
  11. L6:2657 weakly suppresses
  12. L7:833 weakly suppresses
  13. L6:1091 weakly suppresses
  14. L6:1850 weakly suppresses
  15. L6:17 weakly suppresses
  16. L7:795 weakly excites
  17. L7:1470 weakly suppresses
  18. L7:2524 weakly suppresses
  19. L6:1437 weakly excites
  20. L7:1203 weakly excites
  21. L6:16 weakly excites
  22. L7:528 weakly excites
  23. L6:870 weakly excites
  24. L3:1343 weakly suppresses

Feeds later layers

  1. L21:2075 excites
  2. L23:1505 excites
  3. L18:1674 excites
  4. L11:1994 excites
  5. L10:2261 excites
  6. L21:2953 excites
  7. L10:1692 excites
  8. L22:2712 excites
  9. L10:2496 weakly excites
  10. L20:1063 weakly excites
  11. L22:3023 weakly excites
  12. L20:2734 weakly excites
  13. L23:1082 weakly excites
  14. L12:2395 weakly suppresses
  15. L18:2756 weakly excites
  16. L18:895 weakly excites
  17. L19:1522 weakly excites
  18. L26:995 weakly excites
  19. L22:938 weakly excites
  20. L12:993 weakly excites
  21. L9:3070 weakly suppresses
  22. L10:1720 weakly suppresses
  23. L11:549 weakly suppresses
  24. L26:1944 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 -2.19 and -2.001 values between -1.82 and -1.642 values between -0.91 and -0.734 values between -0.73 and -0.5586 values between -0.55 and -0.37620 values between -0.37 and -0.191915 values between -0.19 and -0.011067 values between -0.01 and 0.18121 values between 0.18 and 0.3622 values between 0.36 and 0.5418 values between 0.54 and 0.7216 values between 0.72 and 0.9011 values between 0.90 and 1.093 values between 1.09 and 1.274 values between 1.27 and 1.453 values between 1.45 and 1.631 values between 1.63 and 1.813 values between 1.81 and 1.991 values between 2.18 and 2.361 values between 2.36 and 2.541 values between 2.54 and 2.721 values between 2.72 and 2.902 values between 2.90 and 3.081 values between 3.45 and 3.631 values between 3.81 and 3.991 values between 3.99 and 4.171 values between 4.90 and 5.08offnegativepositiveoften

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. 站点
  2. aint
  3. facilities
  4. 机
  5. 基地
  6. iere
  7. rove
  8. Server
  9. 设施
  10. ivel

Suppresses

  1. шедш
  2. danced
  3. 액
  4. ⇑
  5. ’d
  6. ナー
  7. >b
  8. ós
  9. bic
  10. endant