Neuron Explorer

Qwen3-0.6B

commas in lists of six items

A neuron in layer 11 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 comma that separates items in a list, especially when the list has six things, such as six symptoms or six types. Turning it up made the text drift into lists and numbers, though the link to the theme is loose.

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

    least three of the following six symptoms associated with the anxiety on more days than not in the last six months: restlessness, fatigue, irritability, muscle tension, difficulty sleeping and difficulty concentrating.⏎Obsessive Compulsive Disorder (OCD)⏎

  2. #2 peak

    writing skills. The CD is divided into six sections, each focusing on a different part of speech: nouns/pronouns, verbs, adjectives, adverbs, conjunctions, and prepositions. Each section:⏎- Defines the part of speech

  3. #3 peak

    180, and 200, resulting in stepping angles of 30, 15, 5, 2.5, 2, and 1.8° per step.⏎Step angle = 3

  4. #4 peak

    . According to the online Encyclopedia of Business, 2nd edition, there are six basic types of teams. These include informal, traditional, self-directed, leadership, problem solving and virtual teams. Each specific type of team requires its own individual type

  5. #5 peak

    six grade-appropriate math problems: one each of addition, subtraction, multiplication, division, fractions, and percents. I stressed "grade-appropriate" because single digit

  6. #6 peak

    .⏎UK Politics: The UK’s six main political parties-the Liberal Democrats, the Labour Party, the Conservatives, the Scottish National Party, the Liberal Party of Australia and New Zealand’s Labor Movement – compete for power through

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 am a digital estate, and I have no control over what is stored in my digital estate, but I want to preserve the information I can't, and I am doing this for the purpose of the first phase of the project,

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

    ,500 SMART Boards and interactive projectors have been installed in schools since the Referendum was renewed.

  2. #2 peak

    being.⏎What is blood cholesterol? Also referred to simply as cholesterol, blood cholesterol refers to how much of a lipid (fat) called cholesterol we have per deciliter of blood. Cholesterol is basically

  3. #3 peak

    | M | N | O | P | Q | R | S | T | U | V | W | X | Y | Z | 0-9⏎21 results for Schools: |⏎See previous results⏎See results 1

  4. #4 peak

    for 5-10 days. During this time in the bed bug life cycle, they do not feed but instead digest their meal, mate, and lay eggs. ~ source⏎You may notice a bed bug infestation in your home if

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. L10:559 suppresses
  2. L5:158 weakly suppresses
  3. L5:86 weakly suppresses
  4. L3:1373 weakly excites 'be' after 'to' in infinitives
  5. L10:944 weakly excites
  6. L8:3 weakly excites
  7. L3:50 weakly excites
  8. L1:747 weakly suppresses
  9. L2:86 weakly excites
  10. L6:2596 weakly suppresses
  11. L5:1877 weakly suppresses
  12. L9:1183 weakly suppresses
  13. L9:2421 weakly suppresses
  14. L10:24 weakly suppresses
  15. L8:2264 weakly excites
  16. L5:1359 weakly excites
  17. L7:901 weakly excites
  18. L7:2691 weakly suppresses
  19. L6:2213 weakly suppresses
  20. L4:3051 weakly excites
  21. L7:1491 weakly suppresses
  22. L10:2112 weakly excites
  23. L10:88 weakly excites
  24. L6:2620 weakly suppresses

Feeds later layers

  1. L12:1440 strongly suppresses
  2. L14:1623 strongly suppresses
  3. L13:2423 strongly excites commas in lists of names or items
  4. L13:1941 suppresses
  5. L16:808 excites
  6. L12:1854 excites
  7. L16:2442 suppresses
  8. L14:1215 excites
  9. L12:1977 excites
  10. L15:3020 suppresses
  11. L14:843 excites
  12. L13:564 excites
  13. L16:2880 excites
  14. L12:763 excites
  15. L15:2349 suppresses
  16. L13:1112 excites
  17. L13:899 excites
  18. L15:941 excites
  19. L16:1364 excites
  20. L15:2861 excites 'only' in 'not only ... but'
  21. L16:1809 excites comma inside a list of names or acronyms
  22. L13:1389 excites
  23. L14:1239 excites
  24. L16:2922 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 -0.60 and -0.548 values between -0.54 and -0.4816 values between -0.48 and -0.4143 values between -0.41 and -0.35100 values between -0.35 and -0.28156 values between -0.28 and -0.22301 values between -0.22 and -0.16483 values between -0.16 and -0.09670 values between -0.09 and -0.03768 values between -0.03 and 0.04579 values between 0.04 and 0.10373 values between 0.10 and 0.16187 values between 0.16 and 0.23129 values between 0.23 and 0.2953 values between 0.29 and 0.3523 values between 0.35 and 0.427 values between 0.42 and 0.483 values between 0.48 and 0.552 values between 0.55 and 0.611 values between 0.74 and 0.802 values between 0.80 and 0.871 values between 0.93 and 0.991 values between 1.89 and 1.95offnegativepositiveoften

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. 和
  2. pot
  3. 和技术
  4. และ
  5. 或
  6. 好事
  7. and
  8. ITER
  9. rant
  10. pekt

Suppresses

  1. ability
  2. heimer
  3. ESA
  4. unger
  5. esa
  6. hazi
  7. 的新
  8. umb
  9. ABILITY
  10. anness