Neuron Explorer

Qwen3-0.6B

ampersand in citation author lists

A neuron in layer 15 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 "&" sign in academic citations that join the last two authors, such as "Pyszczynski, Solomon, & Greenberg". It stays mostly quiet elsewhere. Turning it up made the model write lists of items and topics with no clear link to citations.

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

    ifying fears, and creating dependency. Reminding people of their mortality tends to make them hold more closely to traditional culture (Pyszczynski, Solomon, & Greenberg, 2003); this has implications for mental health,

  2. #2 peak

    of trophic levels, can reach very high densities, often have noticeable effects on prey, and are commonly associated with human trade (Suarez, Holway & Ward, 2005; Snyder & Evans, 200

  3. #3 peak

    in comparison with the previous term, mental retardation. An online questionnaire, based on the Multidimensional Attitudes Scale (Findler, Vilchinsky, & Werner, 2007), was

  4. #4 peak

    a range of biodiversity applications such as modelling the distribution of communities, ecological refuges, potential impacts under climate change, and biotic interactions (Araújo & Luoto, 2007; Bradley, 2013

  5. #5 peak

    intention has generally been insignificant among older and younger adults. (Courneya, Estabrooks & Nigg 1997) Unlike attitude, which tends to consistently account

  6. #6 peak

    model, which can be used to predict a species’ potential occurrence across a landscape, have become a key part of ecological research and conservation planning (Guisan & Thuiller, 2005; Thuiller et al., 2

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 value of a person, or a person, in a society, as we go along, with a culture, where we accept, or something, that is, a culture, in which we accept, or, perhaps, the culture,

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

    uo Huang, and Chia-Liang Chiang, "Purification Study of Pervious Concrete Pavbement," International Journal of Engineering and Technology vol. 5, no. 5, pp. 532-5

  2. #2 peak

    ura.⏎Risks of thoracentesis are:⏎Serious complications are uncommon.⏎Broaddus VC, Light RW. Pleural effusion. In: Mason RJ, Murray JF, Broaddus VC, et al., eds. Murray

  3. #3 peak

    in the world to stargaze.⏎"This is where the stars shine - they don't twinkle, they shine," Johnstone says. He spends as much time as possible gazing at the skies: "I haven't discovered anything of importance

  4. #4 peak

    Bible Dictionary, Third Edition,⏎published by Thomas Nelson, 1897. Public Domain, copy freely.⏎Bibliography InformationEaston, Matthew George. "Entry for Tamarisk". "Easton

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. L14:2253 strongly suppresses
  2. L14:2961 excites
  3. L14:837 suppresses
  4. L14:2812 suppresses 'between' after link or relationship nouns
  5. L14:1628 suppresses
  6. L13:706 excites
  7. L14:848 weakly excites
  8. L14:2675 weakly excites
  9. L1:2724 weakly suppresses
  10. L4:833 weakly suppresses
  11. L13:1088 weakly excites
  12. L11:1309 weakly suppresses
  13. L10:335 weakly suppresses commas in catalog, date, and list metadata
  14. L3:1887 weakly excites
  15. L12:1287 weakly excites
  16. L13:2416 weakly excites
  17. L14:2590 weakly suppresses
  18. L10:1580 weakly suppresses
  19. L13:824 weakly suppresses
  20. L10:330 weakly excites
  21. L14:2050 weakly suppresses
  22. L8:471 weakly suppresses
  23. L9:77 weakly suppresses commas in citations and name lists
  24. L6:502 weakly excites

Feeds later layers

  1. L18:862 strongly excites hedges like 'it turns out' and 'if any'
  2. L19:845 strongly excites
  3. L21:344 strongly excites
  4. L20:173 strongly excites
  5. L16:650 strongly suppresses
  6. L16:3060 strongly excites
  7. L26:236 strongly excites
  8. L25:438 strongly excites
  9. L20:276 excites
  10. L24:572 excites
  11. L26:22 suppresses the 'res' fragment inside words
  12. L24:216 excites
  13. L27:286 excites
  14. L24:642 excites
  15. L27:86 suppresses
  16. L18:1404 excites
  17. L24:6 suppresses
  18. L17:449 excites
  19. L23:100 excites comma or full stop ending a clause
  20. L17:735 suppresses
  21. L23:49 excites
  22. L20:1083 excites
  23. L26:2123 excites
  24. L17:1657 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.

2 values between -1.08 and -0.912 values between -0.91 and -0.7514 values between -0.75 and -0.5999 values between -0.59 and -0.42424 values between -0.42 and -0.261253 values between -0.26 and -0.101801 values between -0.10 and 0.06255 values between 0.06 and 0.2334 values between 0.23 and 0.398 values between 0.39 and 0.555 values between 0.55 and 0.723 values between 0.72 and 0.881 values between 1.04 and 1.211 values between 1.37 and 1.531 values between 2.18 and 2.351 values between 2.35 and 2.511 values between 3.00 and 3.161 values between 3.49 and 3.651 values between 3.98 and 4.141 values between 5.28 and 5.44offnegativepositiveoften

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. !,
  4. (),
  5. [],
  6. '/',
  7. !!,
  8. ~,
  9. \",
  10. ",",

Suppresses

  1. `).
  2. `)
  3. =").
  4. `.
  5. ").
  6. ").
  7. ）。
  8. `;⏎
  9. ').
  10. ".