Neuron Explorer

Qwen3-0.6B

year digit 1 and legal words

A neuron in layer 23 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

Mixed neuron. On the positive side it fires on the first digit "1" of years like 1763 or 1545, which are historical dates. On the negative side it fires on legal words such as "attorney", "legal" and "L.P.". When turned down, the model started repeating "law", and when turned up it repeated "subject".

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

    signing of the Treaty of Paris in 1763.⏎Colonial Military Policy⏎Understanding the need for military discipline,

  2. #2 peak

    Accepted by||Catholic Church|⏎|Fifth Council of the Lateran (1512–1517)|⏎|First Vatican Council (1869–1870)|⏎|Convoked by||Paul

  3. #3 peak

    in 1763 through the Peace of Paris, but Spain regained it after Britain's defeat by the American colonies and the subsequent Treaty of Versailles (1

  4. #4 peak

    the Catholic Church⏎The Council of Trent (Latin : Concilium Tridentinum), held between 1545 and 1563 in Trent (or Trento),

  5. #5 peak

    doctrine of the Roman Catholic Church at the Fourth Lateran Council that began on November 11, 1215. The Council of Trent in 1551 confirmed the same doctrine. And it’s the official teaching of the

  6. #6 peak

    0) grants sole trading rights in Hudson Bay drainage basin to the Hudson's Bay Co.⏎By the Treaty of Paris (1763), eastern North America becomes British territory except St-Pierre and M

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 have been representing law firms and law law law law law law law law law law law law law... I have been represented by a law law law law... I have been represented by a law law law... I have been represented by a

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

    you may want to know more about the lifespan of the trust. The Houston living trust lawyers at Mendel Law Firm, L.P. explain when and how a living trust terminates.⏎What Is a Living Trust

  2. #2 peak

    pects can invoke the Sixth Amendment right to counsel during lineups and showups held after they have been arraigned or indicted. They have the right to have an attorney present, but the attorney only participates as an observer.⏎Due process

  3. #3 peak

    to reduce the risk of vehicle fires.⏎Don’t let your losses in a vehicle fire harm you for years to come, know your rights and protect yourself through skilled legal representation.<|endoftext|>In the Vedic tradition, holidays are calculated according to the lunar

  4. #4 peak

    a fairly typical trajectory to the White House, checking off many of the most common boxes for U.S. presidents—he was educated at Harvard Law School and opened a law practice before serving in the military, then became a Congressman, and then Governor of

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. L19:1531 excites
  2. L10:2710 excites
  3. L5:2358 suppresses
  4. L3:2860 suppresses
  5. L3:1101 suppresses
  6. L6:2689 weakly excites
  7. L6:2831 weakly suppresses
  8. L7:1723 weakly suppresses
  9. L22:2542 weakly excites
  10. L9:2970 weakly suppresses
  11. L21:206 weakly suppresses
  12. L11:620 weakly suppresses
  13. L7:2358 weakly suppresses
  14. L10:2187 weakly suppresses
  15. L20:2288 weakly excites
  16. L22:1682 weakly excites
  17. L21:2934 weakly excites
  18. L6:2757 weakly suppresses
  19. L21:2545 weakly suppresses
  20. L7:1482 weakly suppresses
  21. L4:26 weakly excites
  22. L13:965 weakly excites
  23. L6:1481 weakly excites
  24. L4:732 weakly suppresses

Feeds later layers

  1. L25:2899 suppresses
  2. L27:2054 suppresses
  3. L27:1249 suppresses
  4. L26:1398 weakly suppresses
  5. L27:2068 weakly suppresses
  6. L26:2355 weakly suppresses
  7. L25:2516 weakly suppresses
  8. L26:1463 weakly suppresses
  9. L25:1805 weakly suppresses
  10. L27:1119 weakly suppresses
  11. L24:1418 weakly suppresses
  12. L24:588 weakly suppresses
  13. L24:2741 weakly excites
  14. L25:2919 weakly suppresses
  15. L24:470 weakly suppresses
  16. L27:330 weakly suppresses
  17. L25:879 weakly suppresses
  18. L24:1070 weakly excites
  19. L26:2833 weakly suppresses
  20. L24:926 weakly excites
  21. L27:1219 weakly suppresses
  22. L24:2496 weakly excites
  23. L24:2962 weakly excites
  24. L25:289 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 -43.13 and -41.641 values between -23.78 and -22.301 values between -16.34 and -14.861 values between -14.86 and -13.372 values between -10.39 and -8.901 values between -8.90 and -7.424 values between -7.42 and -5.934 values between -5.93 and -4.442 values between -4.44 and -2.9523 values between -2.95 and -1.461304 values between -1.46 and 0.022514 values between 0.02 and 1.5142 values between 1.51 and 3.002 values between 3.00 and 4.491 values between 4.49 and 5.981 values between 7.46 and 8.951 values between 8.95 and 10.441 values between 10.44 and 11.931 values between 11.93 and 13.421 values between 14.90 and 16.39offnegativepositiveoften

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. ode
  2. RNA
  3. 贡
  4. 靴
  5. subj
  6. оф
  7. -sizing
  8. astro
  9. モノ
  10. subject

Suppresses

  1. law
  2. law
  3. 律
  4. Law
  5. 法律
  6. Law
  7. LAW
  8. represented
  9. 法
  10. representation