Neuron Explorer

Qwen3-0.6B

comma after an intro phrase

A neuron in layer 25 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 a comma that ends a short opening phrase and sets up the main sentence, like "To create an appropriate warning letter," in a how-to. Its negative side reacts to the start of words beginning with "Ext", such as "Extremist" or "Extinct". Turning it up gave repeated words like "moon", so the steering is not very informative.

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

    these problem-solving tools in action with a series of 3rd-grade problems based on the Singapore Primary Math series, level 3A. For your reading pleasure, I have translated the problems into the universe of a well-written biography of Ben Franklin

  2. #2 peak

    these threats and violences in multiple and intersecting ways. The deaths of Black women like Breonna Taylor, Sandra Bland, Regis Korchinski-Paquet, and others at the hands of the police cannot be forgotten. We acknowledge

  3. #3 peak

    the game boards.⏎- Attempt to explain to students how to properly pronounce and say dates in English:⏎Once you have got that beast of a task taken care of, simply tell the kids to play "sug

  4. #4 peak

    . But if a predator threatens, eggs may hatch early—and away run the lizards!⏎Ecologist Sean Doody discovered this while in the field with student Phillip Paull. The two were lifting eggs out of a crevice when . . .

  5. #5 peak

    for drug or alcohol abuse, suicidal thoughts or actions, problems at home or work, or deeper social withdrawal.⏎To beat SAD into the ground during the winter, fall,

  6. #6 peak

    apps have made their way onto most electronic devices, you might think there would be dozens of worthwhile products available on the topic designed for students. Think again.⏎Sure, there are plenty of apps devoted to November’s election, there just aren’t that

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: some people might not like the way the process is currently in hand, as it could lead to the finalization of some very influential, very influential, very influential, very influential, very influential, very influential, very influential, very influential,

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

    and who openly condemn whites as evil personified, deserving only death or slavery,” the SPLC said.⏎Are they the guys yelling on street corners?⏎Sometimes. Extremist Hebrew Israelites can be found on street corners literally shouting their doctrine,

  2. #2 peak

    of the different hormonal activities in her body. Also, metabolism is boosted after menopause, thereby leading to hot flashes<|endoftext|>Our newest title⏎Busy with Bugs⏎Extremely Interesting Things

  3. #3 peak

    ?⏎Modern Hebrew is the official language of the State of Israel, while premodern Hebrew is used for prayer or study in Jewish communities around the world today.⏎|Extinct||Mishnaic Hebrew extinct as a spoken language by the 5

  4. #4 peak

    . They have been reintroduced to National Reserves of Mongolia and China⏎CONSERVATION STATUS: Classified as Extinct in the Wild until 2008, but since a series 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. L20:2711 excites
  2. L23:1851 excites
  3. L18:2401 excites
  4. L17:2793 excites
  5. L21:2394 suppresses
  6. L20:2840 suppresses
  7. L22:992 excites
  8. L22:885 suppresses
  9. L19:1316 suppresses
  10. L23:1784 suppresses
  11. L22:1498 weakly excites
  12. L8:2681 weakly excites
  13. L5:26 weakly excites
  14. L23:298 weakly excites
  15. L18:126 weakly suppresses comma joining a full clause
  16. L22:1592 weakly excites
  17. L16:655 weakly excites 'see' cross-references and web boilerplate
  18. L6:1749 weakly excites
  19. L3:2147 weakly excites
  20. L15:152 weakly suppresses
  21. L20:122 weakly excites
  22. L16:348 weakly suppresses
  23. L17:218 weakly excites
  24. L17:845 weakly excites comma after an introductory phrase

Feeds later layers

  1. L27:1269 strongly excites
  2. L26:1080 excites
  3. L26:2360 weakly excites
  4. L27:1504 weakly excites
  5. L26:2198 weakly excites
  6. L26:2663 weakly excites
  7. L27:524 weakly excites
  8. L27:2429 weakly suppresses
  9. L27:2961 weakly suppresses
  10. L26:2537 weakly suppresses
  11. L27:2962 weakly suppresses
  12. L27:2370 weakly suppresses
  13. L26:186 weakly excites
  14. L27:1889 weakly suppresses
  15. L27:2377 weakly excites
  16. L26:1258 weakly suppresses
  17. L27:2167 weakly excites
  18. L27:2457 weakly excites
  19. L27:2502 weakly excites
  20. L26:74 weakly suppresses
  21. L27:2742 weakly excites 'w' inside split words
  22. L27:2666 weakly excites
  23. L27:1116 weakly excites
  24. L26:405 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 -31.25 and -29.861 values between -22.92 and -21.532 values between -17.36 and -15.981 values between -15.98 and -14.591 values between -11.81 and -10.422 values between -10.42 and -9.034 values between -9.03 and -7.644 values between -7.64 and -6.255 values between -6.25 and -4.8710 values between -4.87 and -3.4833 values between -3.48 and -2.0999 values between -2.09 and -0.703386 values between -0.70 and 0.69211 values between 0.69 and 2.0854 values between 2.08 and 3.4722 values between 3.47 and 4.8619 values between 4.86 and 6.2411 values between 6.24 and 7.6316 values between 7.63 and 9.029 values between 9.02 and 10.414 values between 10.41 and 11.803 values between 11.80 and 13.192 values between 13.19 and 14.581 values between 14.58 and 15.974 values between 15.97 and 17.351 values between 17.35 and 18.741 values between 18.74 and 20.131 values between 22.91 and 24.30offnegativepositiveoften

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. here
  2. here
  3. Here
  4. 在这里
  5. 这里
  6. Here
  7. 这里是
  8. HERE
  9. ici
  10. aquí

Suppresses

  1. ambda
  2. ourt
  3. processo
  4. 刘海
  5. FragmentManager
  6. ought
  7. możli
  8. 铿
  9. .getJSONObject
  10. UILD