Neuron Explorer

Qwen3-0.6B

end of a quoted or bracketed phrase

A neuron in layer 0 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 at the spot right after a quoted phrase or a bracketed name ends, where a closing quote or bracket is expected. Examples include the end of a speaker's quote or a name in square brackets. Turning it up made the text repeat simple words like "still", which does not clearly relate to the pattern.

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

    as institutions for rational choice and behaviouralism is a loss and therefore needs to be accommodated. This is where institutionalism comes in, which seeks to “[explain] the decisions that the governments make” – be it their values, rules, incentives

  2. #2 peak

    -ball, scientists say that would actually increase the intensity of the sun’s radiation, contrary to the film. Rather than Australia becoming a frozen tundra, “[we] would be fried,” according to Kusenko.⏎And then, when our heroes

  3. #3 peak

    possessing an almost ‘amphibious’ quality.⏎British-American classical scholar Anthony Long, author of Greek Models of Mind and Self, writes: “[Homer] Gives a marvelously detailed and realistic anatomy of the feeling… In his description both

  4. #4 peak

    conservation strategies 4, 5, 6. “The effects of tropical deforestation on climate go well beyond carbon,” says Professor Deborah Lawrence, “[it] causes warming locally, regionally, and. IDENT

  5. #5 peak

    ities to directly convert electrical energy to thrust without the need to expel any propellant,” explains Hacked.com. “[Harold G. White, a scientist] proposes that the EM Drive’s thrust is due to virtual particles in the quantum vacuum

  6. #6 peak

    freedom provided by FDM has helped us improve the performance and reduced the cost of many of our products," says McCabe. "The [Fortus] FDM system produces parts with the strength and accuracy to survive and provide data while

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 trying to write this in a way to try to write this in a way to think, and to write this in a way to try to think, and to write this in a way to try to write. It seems to

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

    be taken into consideration in all circumstances.⏎- ^ According to the IEC there is no clear boundary between medium and high voltage.

  2. #2 peak

    more from the Hundred Years' War than England did. During the war, France was ravaged by civil strife as its population, economy and resources depleted.⏎- ^ Swyrich, Archive materials⏎This page was last modified on 5 January

  3. #3 peak

    , a French astrophysicist at NASA Langley.⏎"And you think, 'OK, what I am looking at — it's a<|endoftext|>An ultra-Orthodox Jewish circumcision practice in which the circumciser places his mouth on a

  4. #4 peak

    province of Segovia, in 1559; died at Madrid, 27 March, 1625. He was a great-grandson of the Tordesillas who was put to death by the Comuneros

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

None: this is the first layer.

Feeds later layers

  1. L10:811 weakly excites
  2. L2:627 weakly suppresses
  3. L3:2353 weakly suppresses
  4. L17:237 weakly suppresses
  5. L10:2185 weakly suppresses
  6. L3:1807 weakly suppresses
  7. L22:441 weakly suppresses
  8. L9:1727 weakly suppresses
  9. L13:2390 weakly excites limits, thresholds and how long it takes
  10. L4:1153 weakly suppresses
  11. L8:2185 weakly excites
  12. L9:1018 weakly excites
  13. L2:505 weakly suppresses
  14. L9:2322 weakly excites
  15. L2:1438 weakly excites
  16. L8:1761 weakly suppresses
  17. L7:1662 weakly excites
  18. L17:244 weakly excites
  19. L5:567 weakly suppresses
  20. L7:1342 weakly excites
  21. L22:364 weakly suppresses
  22. L17:1732 weakly suppresses
  23. L24:669 weakly suppresses
  24. L2:396 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 -0.47 and -0.441 values between -0.44 and -0.411 values between -0.41 and -0.384 values between -0.38 and -0.364 values between -0.36 and -0.333 values between -0.30 and -0.279 values between -0.27 and -0.2410 values between -0.24 and -0.2213 values between -0.22 and -0.1921 values between -0.19 and -0.1638 values between -0.16 and -0.1347 values between -0.13 and -0.1089 values between -0.10 and -0.08173 values between -0.08 and -0.05456 values between -0.05 and -0.021546 values between -0.02 and 0.01704 values between 0.01 and 0.04335 values between 0.04 and 0.07190 values between 0.07 and 0.09115 values between 0.09 and 0.1259 values between 0.12 and 0.1525 values between 0.15 and 0.1822 values between 0.18 and 0.2016 values between 0.20 and 0.238 values between 0.23 and 0.262 values between 0.26 and 0.296 values between 0.29 and 0.323 values between 0.32 and 0.341 values between 0.37 and 0.402 values between 0.40 and 0.431 values between 0.43 and 0.462 values between 0.62 and 0.65offnegativepositiveoften

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. enced
  2. endez
  3. edin
  4. voke
  5. zych
  6. raction
  7. ed
  8. edes
  9. plet
  10. _RDONLY

Suppresses

  1. 品类
  2. argon
  3. population
  4. Evaluate
  5. catastrophe
  6. darn
  7. expressing
  8. urn
  9. 复杂的
  10. turbulence