Neuron Explorer

Qwen3-0.6B

"a" after give or get verbs

A neuron in layer 21 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 "a" when it follows verbs like "give", "get" or "gave", as in "give him a chance" or "gave it a second thought". It is pushed negative by the word "gain" in phrases like "gain access". Turning it down made the model loop on the word "access".

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

    ing individual is given a warning letter to provide him with ample time to explain his side on the matter as part of the due process. This will give him/her a chance to correct himself/herself before being sanctioned.⏎Actions that need disciplinary action

  2. #2 peak

    high University in Bethlehem, Pa.⏎"I never gave evolution a second thought, and certainly saw no⏎conflict with Church teaching," said Be

  3. #3 peak

    way to sate your cravings than the sugary stuff, nutritionist Lauren Slayton’s explanation of why diet coke isn’t better than coke might give you a reality check.⏎5. Also, did you know that Diet Coke explodes when you

  4. #4 peak

    will die.”⏎Mr O’Connor suggested the best thing people could do to lend the birds a hand was to place a dish of water in the shade.⏎“So that the birds

  5. #5 peak

    — The Founding Fathers didn’t considered the Declaration of independence a piece of propaganda.⏎— Thomas Jefferson didn’t say a word at the convention. And never gave a single speech as president. He would rather write than speak.⏎— Jefferson got the

  6. #6 peak

    . It was soon clear, however, that the young man was unsuited for the work and he agreed with his father to study law, to give him a steady income should writing – already his chosen path – fail him.⏎Literature,

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: access to access. Access has access. Access has access access. Access has access access access access... and access access access access access access access access access access... access access access access access access access access access access access access access access access access

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

    funds⏎Another excellent example of multi-factor authentication is using an SMS, voice message, or email to send a code the user can input along with a password to gain access. In this case, you’re combining:⏎- Knowledge – the user must

  2. #2 peak

    revenue. In addition, healthcare organizations may be subject to fines from regulatory agencies for violating data protection laws.⏎- Loss of valuable PHI: If a hacker manages to gain access to the organization’s network, they could steal or delete sensitive data. Additionally

  3. #3 peak

    don’t have anything worth stealing.”⏎Short for malicious software, malware refers to any program that is introduced into the target's computer with the purpose of causing damage, gaining unauthorized access, or displaying unwanted advertising. Under the malware umbrella, other types of

  4. #4 peak

    930⏎Our editors will review what you’ve submitted and determine whether to revise the article.Join Britannica's Publishing Partner Program and our community of experts to gain a global audience for your work!⏎One of the youngest of the world’s major

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. L8:2673 suppresses
  2. L14:1684 excites
  3. L20:2118 excites
  4. L20:2854 suppresses 'official' and government institution words
  5. L16:1860 suppresses
  6. L17:2009 suppresses
  7. L17:3015 suppresses
  8. L8:2741 weakly suppresses
  9. L4:3023 weakly excites
  10. L14:2868 weakly excites
  11. L16:2855 weakly excites
  12. L17:1370 weakly excites
  13. L19:1877 weakly excites
  14. L5:2042 weakly excites
  15. L4:1334 weakly excites
  16. L19:2010 weakly suppresses
  17. L16:804 weakly suppresses
  18. L17:1479 weakly suppresses
  19. L14:881 weakly suppresses
  20. L0:1972 weakly excites
  21. L20:1318 weakly suppresses
  22. L9:2505 weakly suppresses
  23. L3:2690 weakly suppresses
  24. L3:1218 weakly suppresses

Feeds later layers

  1. L27:1707 suppresses
  2. L22:2752 suppresses
  3. L25:2403 suppresses
  4. L22:3055 suppresses
  5. L24:2746 suppresses
  6. L25:1872 suppresses
  7. L24:1250 weakly suppresses mixed: 'New' in place names, square roots
  8. L22:2861 weakly suppresses
  9. L24:1599 weakly suppresses "of" in "of use/concern/interest" phrases
  10. L25:764 weakly suppresses
  11. L23:2449 weakly suppresses
  12. L22:2509 weakly suppresses
  13. L24:2925 weakly suppresses
  14. L22:3009 weakly suppresses
  15. L27:1296 weakly suppresses
  16. L22:1647 weakly suppresses
  17. L27:2065 weakly suppresses
  18. L26:762 weakly suppresses
  19. L23:2867 weakly suppresses
  20. L22:2266 weakly suppresses
  21. L23:2622 weakly suppresses
  22. L24:2964 weakly suppresses
  23. L22:2587 weakly suppresses
  24. L22:670 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.

2 values between -31.52 and -30.421 values between -26.01 and -24.911 values between -20.51 and -19.411 values between -19.41 and -18.301 values between -15.00 and -13.901 values between -13.90 and -12.801 values between -12.80 and -11.702 values between -10.60 and -9.501 values between -9.50 and -8.404 values between -8.40 and -7.292 values between -7.29 and -6.197 values between -6.19 and -5.092 values between -5.09 and -3.9916 values between -3.99 and -2.8936 values between -2.89 and -1.79104 values between -1.79 and -0.693007 values between -0.69 and 0.41689 values between 0.41 and 1.5116 values between 1.51 and 2.623 values between 2.62 and 3.723 values between 3.72 and 4.822 values between 4.82 and 5.922 values between 5.92 and 7.022 values between 8.12 and 9.221 values between 9.22 and 10.321 values between 11.42 and 12.52offnegativepositiveoften

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. azar
  2. odings
  3. zie
  4. combe
  5. vant
  6. det
  7. уй
  8. あります
  9. ffe
  10. 您的孩子

Suppresses

  1. access
  2. access
  3. Access
  4. Access
  5. _access
  6. (access
  7. acesso
  8. acceso
  9. .access
  10. 权限