Neuron Explorer

Qwen3-0.6B

comma opening a list of parallel items

A neuron in layer 19 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 reacts to a comma in the middle of a sentence that sets up more items or a second part, such as 'X, Y, or Z' or 'not only A, but also B'. When it was turned up, the model's text tended to fall into long comma-separated lists of similar things.

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

    improvements ensure the app’s quality and usability. Regular updates and bug fixes based on user feedback contribute to long-term success. Lastly, keeping an eye on industry trends, exploring new technologies, and participating in developer communities foster growth and innovation

  2. #2 peak

    associated with loss of appetite in the elderly.⏎- Hepatitis or chronic liver disease. Loss of appetite is known as one of the first symptoms associated with both hepatitis, or inflammation of the liver, and chronic liver disease.⏎- Depression. Psychological issues

  3. #3 peak

    the intensive care unit (ICU) who have had very big operations, such as major organ reconstructions. “There is a race between making the patient better, fixing the bleeding, and

  4. #4 peak

    is inserted through a small incision in your belly. This procedure is generally performed under general anesthesia in a hospital.⏎- MRI. This test uses a large magnet, radio waves, and a computer to make 2-D images of an internal

  5. #5 peak

    around the world. This will not only benefit biodiversity, but will also help build local resilience to the increasing impacts of climate change. Three

  6. #6 peak

    From the Internet<|endoftext|>Fraud is a term that is used to describe deception perpetrated in order to achieve some kind of goal or benefit. Fraud can be either civil, criminal, or both, with civil fraud being punishable by the assignment of monetary damages

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: and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and

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

    soil moist all the way through. It also controls the growth of weeds. Mulch, therefore, fulfils the dual purpose of keeping the weeds away as well as providing nutrients to the soil.⏎But what exactly is mulch?

  2. #2 peak

    amplified by two interrelated aspects of 21st-century life: firstly, the “post-truth” era where the opinions of experts are viewed with skepticism, where everyone’s opinions are equally valid and opposing views (and evidence) dismissed as

  3. #3 peak

    and mouse ports were introduced with it. PS/2 ports connect the keyboard and mouse to a computer and are usually color-coded on today's systems - purple for keyboards and green for mice. Most desktop

  4. #4 peak

    . The type of relationship⏎that exits is indicated by the slope of the diagram.⏎A scatter diagram is composed of a horizontal axis containing the measured values of one variable and a vertical axis representing the measurements of the other variable.⏎The purpose of the

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. L18:2516 strongly excites
  2. L18:1549 strongly excites
  3. L18:1402 strongly excites
  4. L17:2155 excites
  5. L17:1738 excites 'and' closing a list of items
  6. L17:2332 suppresses 'and' joining two clauses
  7. L16:473 excites
  8. L16:1766 excites
  9. L17:1134 suppresses
  10. L18:2118 excites
  11. L17:483 suppresses
  12. L15:2900 excites
  13. L16:808 suppresses
  14. L13:1066 suppresses
  15. L16:1119 suppresses
  16. L16:2183 excites
  17. L13:1296 excites
  18. L15:2500 excites 'and' closing a list of items
  19. L18:1693 excites
  20. L16:548 excites
  21. L13:1626 excites
  22. L13:557 excites
  23. L14:2459 suppresses
  24. L16:2411 suppresses

Feeds later layers

  1. L24:205 strongly suppresses
  2. L20:1009 strongly suppresses
  3. L25:6 strongly excites
  4. L26:347 strongly suppresses the 're' in self-reliant and similar
  5. L22:632 strongly suppresses
  6. L27:289 strongly suppresses
  7. L23:379 strongly suppresses
  8. L26:65 strongly excites
  9. L21:739 strongly suppresses 'both X and Y' pairing words
  10. L20:75 excites
  11. L25:801 suppresses
  12. L25:583 suppresses
  13. L20:560 suppresses
  14. L20:610 suppresses
  15. L21:154 suppresses
  16. L26:859 suppresses
  17. L25:1270 excites
  18. L26:74 excites
  19. L27:1902 excites
  20. L20:68 excites
  21. L22:791 suppresses
  22. L21:1059 weakly excites
  23. L26:2473 weakly suppresses
  24. L27:2212 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.

3 values between -2.75 and -2.432 values between -2.43 and -2.103 values between -2.10 and -1.784 values between -1.78 and -1.469 values between -1.46 and -1.1343 values between -1.13 and -0.8191 values between -0.81 and -0.49425 values between -0.49 and -0.172643 values between -0.17 and 0.16450 values between 0.16 and 0.4881 values between 0.48 and 0.8033 values between 0.80 and 1.1312 values between 1.13 and 1.4513 values between 1.45 and 1.7714 values between 1.77 and 2.0916 values between 2.09 and 2.425 values between 2.42 and 2.744 values between 2.74 and 3.068 values between 3.06 and 3.398 values between 3.39 and 3.719 values between 3.71 and 4.034 values between 4.03 and 4.352 values between 4.35 and 4.684 values between 4.68 and 5.001 values between 5.00 and 5.324 values between 5.32 and 5.654 values between 5.65 and 5.973 values between 5.97 and 6.291 values between 6.29 and 6.611 values between 6.61 and 6.942 values between 6.94 and 7.261 values between 7.26 and 7.583 values between 8.23 and 8.551 values between 9.52 and 9.841 values between 9.84 and 10.16offnegativepositiveoften

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. ropol
  2. eled
  3. ctrine
  4. ellar
  5. ahl
  6. STRUCTOR
  7. acular
  8. uele
  9. izabeth
  10. iances

Suppresses

  1. and
  2. 和
  3. và
  4. 和个人
  5. 和
  6. และ
  7. 和社会
  8. и
  9. 性和
  10. and