Neuron Explorer

Qwen3-0.6B

unclear: hyphen in COVID-19

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 is weak and mixed. Its clearest trigger is the dash in 'COVID-19', and it also fires on a few odd word joins. Turning it up gave writing that had nothing to do with this.

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

    be coincidental.⏎You can find the rest of the article here:⏎Broadband? Here is a perspective on broadband from the FCC⏎THE CONSENSUS UNDERSTANDING OF THE FCC’S ROLE WITH RESPECT TO BROADBAND⏎A

  2. #2 peak

    related to COVID-19.⏎A major social determinant worsening these health inequities for patients with

  3. #3 peak

    progress in the fight against COVID-19. However, if you feel constant anxiety, thoughts about the pandemic interfere

  4. #4 peak

    even more hesitant to leave their homes to go to the pharmacy. But it is particularly important that such higher-risk patients maintain adherence to reduce their vulnerability to COVID-19.⏎- Public mail carriers will typically leave packages in a mailbox or at

  5. #5 peak

    ica-born writer Jean Rhys, best known for her novel “Wide Sargasso Sea” — a creatively daring, strongly feminist, and brazenly anti-colonial counter to Charlotte Brontë's “Jane Eyre” — is

  6. #6 peak

    pandemics such as COVID-19. Suppression can work if the outbreak is small and you’re quick

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 am a student who is a student, and I am a student who is a student. " " " " " " " " " " " " " " " " " " " " " " " " " " " "

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

    ceremony of "opening of the mouth". This ceremony was thought to restore the mummy's ability to see, breathe, eat and drink.⏎- The coffin was

  2. #2 peak

    at about 68 degrees latitude.<|endoftext|>Just months after Germany’s chancellor was sent packing by Canada’s Prime Minister when he came seeking natural gas, with a jibe about there being no market for the stuff in Europe, the first shipment from

  3. #3 peak

    retreat after a severe mauling. The leader of the Coorg resistance at was 'Madanta'

  4. #4 peak

    indoor plants alone, while others wouldn’t notice an indoor jungle.⏎Obviously, my home is full to bursting with houseplants, and I have two young children that merrily ignore them. I’m lucky if my daughter makes a passing comment about

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. L0:286 weakly excites
  2. L0:626 weakly excites
  3. L0:1869 weakly suppresses
  4. L0:279 weakly suppresses
  5. L0:230 weakly excites
  6. L0:63 weakly excites
  7. L0:2744 weakly suppresses
  8. L0:426 weakly excites
  9. L0:1174 weakly excites
  10. L0:1790 weakly excites
  11. L0:2654 weakly excites
  12. L0:1534 weakly suppresses
  13. L0:2172 weakly suppresses
  14. L0:793 weakly excites
  15. L0:976 weakly excites
  16. L0:1868 weakly suppresses
  17. L0:427 weakly suppresses
  18. L0:504 weakly suppresses
  19. L0:540 weakly excites
  20. L0:227 weakly suppresses line break after a title
  21. L0:365 weakly excites
  22. L0:31 weakly suppresses
  23. L0:490 weakly suppresses
  24. L0:2603 weakly excites

Feeds later layers

  1. L3:18 suppresses
  2. L3:21 excites
  3. L20:1554 weakly suppresses
  4. L15:1860 weakly suppresses
  5. L25:17 weakly suppresses
  6. L21:70 weakly suppresses
  7. L14:140 weakly suppresses
  8. L16:946 weakly suppresses the word 'late' in time phrases
  9. L16:1674 weakly suppresses
  10. L18:265 weakly suppresses first half of institution names
  11. L18:651 weakly excites
  12. L3:619 weakly excites
  13. L12:1329 weakly suppresses
  14. L25:2282 weakly excites
  15. L9:909 weakly excites
  16. L6:481 weakly excites
  17. L20:535 weakly suppresses
  18. L27:1872 weakly suppresses
  19. L17:2 weakly suppresses
  20. L20:2120 weakly excites "so" opening a purpose clause
  21. L12:1458 weakly suppresses
  22. L25:30 weakly suppresses
  23. L23:878 weakly suppresses
  24. L24:1704 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 -2.13 and -2.052 values between -1.90 and -1.831 values between -1.83 and -1.751 values between -1.67 and -1.601 values between -1.52 and -1.451 values between -1.45 and -1.373 values between -1.30 and -1.223 values between -1.22 and -1.143 values between -1.14 and -1.074 values between -1.07 and -0.995 values between -0.99 and -0.922 values between -0.92 and -0.847 values between -0.84 and -0.764 values between -0.76 and -0.698 values between -0.69 and -0.6110 values between -0.61 and -0.5413 values between -0.54 and -0.4620 values between -0.46 and -0.3921 values between -0.39 and -0.3137 values between -0.31 and -0.2376 values between -0.23 and -0.16280 values between -0.16 and -0.081064 values between -0.08 and -0.011657 values between -0.01 and 0.07461 values between 0.07 and 0.15134 values between 0.15 and 0.2243 values between 0.22 and 0.3020 values between 0.30 and 0.3710 values between 0.37 and 0.456 values between 0.45 and 0.534 values between 0.53 and 0.603 values between 0.60 and 0.682 values between 0.75 and 0.831 values between 0.83 and 0.90offnegativepositiveoften

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. 赍
  2. Custom
  3. ilian
  4. Custom
  5. Globe
  6. نان
  7. custom
  8. (Filter
  9. indication
  10. :y

Suppresses

  1. 静态
  2. 册
  3. incipal
  4. itra
  5. avel
  6. uous
  7. ря
  8. основном
  9. 幔
  10. ($__