Neuron Explorer

Qwen3-0.6B

opening bracket that adds a synonym

A neuron in layer 17 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 opening bracket in phrases that add an explanation or another name, such as "motor fuel (or gas)" or "kefir (fermented milk)." Turning it up or down changed the model's writing in ways that did not seem to relate to brackets.

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

    falling prey to malware — including spyware — is by adhering to several best practices: Install a software or hardware firewall that can detect and selectively block incoming traffic (such as attacks from hackers via your broadband connection to the Internet) and outgoing traffic

  2. #2 peak

    this acceleration in prices. We divide the “basket” of goods and services that make up the CPI into twenty-three items—such as medical care, motor fuel (or gas), apparel, education, and rent—and examine the change in each item

  3. #3 peak

    , usually aged 40 or older, experiences any of these symptoms on a regular (usually daily) basis: eyes that feel dry, moist or watery (yes, Dry Eye Syndrome can sometimes result in eyes that overflow with tears); eyes

  4. #4 peak

    the best available top probiotic foods.⏎Probiotic Drinks List⏎Probiotic drinks include:-⏎– kefir (fermented milk)⏎– water kefir (fermented fruit juice / sugar water

  5. #5 peak

    for its defense.⏎As nuclear policy evolved during the Eisenhower Administration, three factors combined to produce a new American reliance on nuclear weapons: pressure to control the federal budget (the “bigger bang” argument); competition as each branch of the American military

  6. #6 peak

    fun—and tasty!—way to make the parts of the cell more concrete for students is to create edible cells. Using materials such as Jell-O® (to represent cytoplasm) and sprinkles (to represent ribosomes) will not

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: the first time I saw the movie, the story is about a person's dream, the hope that the person has been dreaming about, the hope that the person has been dreaming about, the hope that the person will achieve that dream, the

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

    ranged from 26 to 48 feet (8 to 14.7 meters), and they were between 8 and 16 feet (2.5 and 5 m) wide.⏎The other two ships the researchers examined

  2. #2 peak

    or plants that has characteristics of both parents; hybrid.⏎- A mule is a cross between a horse and a donkey.⏎part of speech: verb⏎inflections: crosses

  3. #3 peak

    to 12 feet (0.61 m to 3.66 m) in width, and 8 feet to 12 feet (2.44 m to 3.66 m) in height. These

  4. #4 peak

    and safety.⏎What is a Risk Assessment?⏎A risk assessment is a very thorough examination of a job, process or task within the workplace, in order to determine what potential risks might occur within that environment. Its intention is to discover any potential causes

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. L15:389 excites
  2. L16:1646 excites
  3. L15:2000 excites
  4. L16:2384 excites
  5. L16:2665 excites
  6. L16:1977 excites
  7. L14:1076 excites
  8. L15:2056 suppresses
  9. L16:2247 suppresses
  10. L16:521 excites
  11. L14:393 excites
  12. L16:717 suppresses
  13. L16:826 suppresses
  14. L16:2251 excites
  15. L16:2327 excites
  16. L16:2200 suppresses
  17. L16:2880 suppresses
  18. L16:2021 suppresses
  19. L16:2457 excites
  20. L15:1783 suppresses
  21. L15:816 excites
  22. L15:3020 suppresses
  23. L15:2190 excites
  24. L15:2277 suppresses

Feeds later layers

  1. L18:2118 strongly suppresses
  2. L20:610 strongly suppresses
  3. L18:1403 strongly suppresses
  4. L18:2838 suppresses
  5. L19:1605 suppresses
  6. L20:156 suppresses
  7. L19:1089 suppresses
  8. L20:1083 suppresses
  9. L19:867 suppresses
  10. L20:460 excites
  11. L19:980 excites
  12. L19:1820 suppresses
  13. L18:1179 excites
  14. L19:2057 suppresses
  15. L21:739 suppresses 'both X and Y' pairing words
  16. L21:301 excites
  17. L18:593 excites
  18. L18:1965 suppresses
  19. L18:2663 suppresses comma ending a clause before an action
  20. L19:1369 suppresses
  21. L19:210 excites
  22. L20:1312 suppresses
  23. L18:670 excites
  24. L19:1262 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 -3.39 and -3.121 values between -3.12 and -2.854 values between -2.85 and -2.585 values between -2.58 and -2.317 values between -2.31 and -2.0422 values between -2.04 and -1.7836 values between -1.78 and -1.5167 values between -1.51 and -1.24103 values between -1.24 and -0.97182 values between -0.97 and -0.70299 values between -0.70 and -0.43444 values between -0.43 and -0.16826 values between -0.16 and 0.11585 values between 0.11 and 0.38427 values between 0.38 and 0.65297 values between 0.65 and 0.92217 values between 0.92 and 1.18122 values between 1.18 and 1.4577 values between 1.45 and 1.7244 values between 1.72 and 1.9941 values between 1.99 and 2.2628 values between 2.26 and 2.5324 values between 2.53 and 2.8010 values between 2.80 and 3.075 values between 3.07 and 3.347 values between 3.34 and 3.614 values between 3.61 and 3.875 values between 3.87 and 4.143 values between 4.14 and 4.412 values between 4.41 and 4.681 values between 4.68 and 4.953 values between 4.95 and 5.223 values between 5.49 and 5.761 values between 5.76 and 6.031 values between 6.03 and 6.302 values between 6.56 and 6.831 values between 7.10 and 7.37offnegativepositiveoften

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. */⏎⏎
  2. »⏎⏎
  3. -elect
  4. ''⏎
  5. ночь
  6. ándose
  7. 」⏎⏎
  8. mell
  9. '''⏎⏎
  10. allocating

Suppresses

  1. ;
  2. <|endoftext|>
  3. ;⏎⏎⏎
  4. ;⏎⏎⏎⏎
  5. ; ⏎
  6. ;⏎
  7. ;
  8. ; ⏎
  9. \Persistence
  10. , ⏎