Neuron Explorer

Qwen3-0.6B

capital P starting a name

A neuron in layer 27 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the capital letter P that starts a name, like 'P' in Ptolemy or Pius. When turned up, the model wrote the word 'her' over and over, and when turned down it wrote 'inheritance' over and over.

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

    .⏎Cleopatra, born in 69 B.C., was made Cleopatra VII, queen of Egypt, upon the death of her father, Ptolemy XII, in 51 B.C. Her brother was

  2. #2 peak

    1–52) were overseen by Pope Julius III and the seventeenth to twenty-fifth sessions (1562–63) by Pope Pius IV.⏎The consequences

  3. #3 peak

    Roman civil war spilled into Egypt. Pompey the Great, defeated by Julius Caesar in Greece, fled to Egypt seeking solace but was immediately murdered by agents of Ptolemy XIII. Caesar arrived in Alexandria soon after and, finding his enemy

  4. #4 peak

    . Thomas Pichler from the University of Vienna report a method for the bulk production of long

  5. #5 peak

    Pope Pius IX define the Immaculate Conception when some great medieval theologians had

  6. #6 peak

    the People can do to remedy their unconstitutional decisions.⏎This course is important and timely for all citizens, students, and homeschoolers. Learn about judicial attacks on the Pledge of Allegiance, the Ten Commandments, the Boy Scouts, the institution

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: inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance inheritance

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

    , additional physical abnormalities. LEOPARD syndrome is an autosomal dominant genetic disorder.⏎LEOPARD syndrome and Noonan syndrome are both caused by mutations in the PTPN11 and RAF1 genes.⏎The symptoms and physical characteristics associated with

  2. #2 peak

    result was the growth of two human embryonic stem cell lines that lack PIG-A, and therefore do not contain any proteins known as glycosylph

  3. #3 peak

    -eating dinosaur, Protoceratops andrewsi, was active both day and night.⏎Nighttime was, in fact, the right time for some pterosaurs, including a fish-eating species and a filter feeder that probably lived

  4. #4 peak

    these plants and animals that are at bottom of the Antarctic food chain to survive. This in turn endangers the population of species higher up the food chain, like penguins and seals.⏎Scientists predict Adélie penguins could become extinct by

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. L21:2745 excites
  2. L23:1099 excites
  3. L24:2165 excites
  4. L24:1460 excites
  5. L22:979 excites
  6. L17:214 excites
  7. L20:322 suppresses
  8. L2:1554 excites
  9. L9:2911 weakly suppresses
  10. L0:497 weakly excites
  11. L22:1780 weakly excites
  12. L26:1770 weakly excites
  13. L17:958 weakly suppresses
  14. L23:1489 weakly excites
  15. L4:1457 weakly excites
  16. L21:346 weakly suppresses
  17. L7:520 weakly excites
  18. L21:1535 weakly suppresses 'Old' as part of a name or phrase
  19. L18:2453 weakly suppresses
  20. L17:2742 weakly excites
  21. L1:1757 weakly suppresses
  22. L24:2710 weakly suppresses
  23. L17:980 weakly suppresses
  24. L24:8 weakly suppresses

Feeds later layers

None: this is the last layer.

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 -31.16 and -29.691 values between -25.30 and -23.843 values between -15.06 and -13.591 values between -13.59 and -12.134 values between -10.67 and -9.204 values between -9.20 and -7.745 values between -7.74 and -6.275 values between -6.27 and -4.8110 values between -4.81 and -3.3511 values between -3.35 and -1.88234 values between -1.88 and -0.423554 values between -0.42 and 1.0543 values between 1.05 and 2.5112 values between 2.51 and 3.975 values between 3.97 and 5.442 values between 5.44 and 6.904 values between 6.90 and 8.362 values between 8.36 and 9.833 values between 11.29 and 12.752 values between 14.22 and 15.681 values between 15.68 and 17.151 values between 25.93 and 27.39offnegativepositiveoften

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. her
  2. 她的
  3. 她
  4. 她在
  5. Her
  6. hers
  7. 让她
  8. Her
  9. herself
  10. 对她

Suppresses

  1. 继承
  2. inherit
  3. inheritance
  4. inherited
  5. inherits
  6. 传承
  7. inherit
  8. 子孙
  9. heritance
  10. inher