Neuron Explorer

Qwen3-0.6B

"can" after a noun subject

A neuron in layer 20 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 word "can" when it comes right after a noun or pronoun that is the subject of the sentence, such as "the directions can" or "it can". It turns negative on words like "parts" or "it" that sit just before a verb. The steering results were mixed, so the link to the outputs is weak.

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

    Association (ADA) seal ensures that it has met standards of safety and effectiveness. When choosing a kit, make sure to read the directions carefully, as the directions can stain clothing and fabrics.⏎There are a few groups of people who should approach this

  2. #2 peak

    can cause bloating.⏎What Is Buckwheat?⏎As one of the lesser known grains out there, buckwheat (and what it is good for) can often be a point of confusion for many. So, before we get any further

  3. #3 peak

    the standard radio option in Tesla cars.⏎“AM radio stations are accessible through our internet radio service in Model X. Because AM audio quality can be very poor, we offer internet radio to give<|endoftext|>A New Brunswick Ghost Story

  4. #4 peak

    navigation in which the bearing is the angle between a line that is used to connect two points. While the navigation bearing is a very important one, the mechanical bearing can be equally important.⏎Mechanical bearings are very different from the navigational bearings that

  5. #5 peak

    Heat transfer is taking place.⏎The picture here is an indoor A-coil. This one is uncased and is built directly into the plenum. Indoor coils can also be purchased built into a case, which installs directly to your air handler or

  6. #6 peak

    a patient’s skin, eyes, lungs, heart, and blood vessels.⏎RA Risk Factors⏎The Mayo Clinic lists the following RA risk factors. While this list can be very useful in helping individuals determine if they are at risk for RA, it

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 right to the rightest possible can be the right of the rightest possible. Can you explain the possible rightings that can be given to the rightest possible, and how the rightest can be the rightest can be the right

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

    For example, using the correct type of filter or the right belt size for the blower motor is essential for the system to function effectively. Substandard or incorrect parts can lead to inefficiencies and even damage the system over time

  2. #2 peak

    often lead to greater mastery of skills, as repetition and practice can lead to improvement over time.⏎However, there are also several disadvantages to hard work. First, it can be time-consuming, leading to burnout and exhaustion. Second, it may

  3. #3 peak

    of injury in a fire or other accident.⏎- Install aftermarket parts correctly. Incorrectly-installed cruise control, stereos, and other aftermarket parts may pose electrical risks that can start a fire. Make sure these parts are installed

  4. #4 peak

    umatic fever can lead to painful and inflamed joints, a specific type of rash, or heart valve damage.⏎Strep throat is most common in children, but it affects people of all ages. If you or your child has signs or symptoms of

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. L19:2560 suppresses
  2. L18:2974 suppresses
  3. L19:285 excites
  4. L19:1862 suppresses
  5. L19:1700 suppresses
  6. L19:2896 excites
  7. L15:322 suppresses
  8. L17:730 excites
  9. L19:2607 excites
  10. L18:2671 excites
  11. L19:2180 suppresses
  12. L18:504 weakly suppresses
  13. L14:685 weakly suppresses
  14. L14:1206 weakly excites
  15. L11:346 weakly suppresses
  16. L14:1347 weakly suppresses
  17. L8:102 weakly excites
  18. L19:2157 weakly suppresses
  19. L18:1832 weakly excites
  20. L8:1370 weakly suppresses
  21. L5:1401 weakly excites
  22. L14:1501 weakly suppresses
  23. L17:1118 weakly suppresses mid-clause words in technical prose
  24. L5:2846 weakly suppresses

Feeds later layers

  1. L27:774 strongly suppresses
  2. L24:766 strongly suppresses
  3. L24:789 strongly suppresses
  4. L22:1422 suppresses
  5. L25:124 excites
  6. L25:1821 weakly suppresses
  7. L23:1193 weakly suppresses
  8. L22:2387 weakly excites
  9. L21:897 weakly excites
  10. L21:953 weakly excites
  11. L22:630 weakly excites
  12. L22:2192 weakly suppresses
  13. L27:212 weakly excites
  14. L21:1435 weakly excites
  15. L22:981 weakly suppresses
  16. L23:469 weakly excites
  17. L22:716 weakly excites
  18. L21:123 weakly excites
  19. L25:752 weakly suppresses
  20. L22:289 weakly suppresses
  21. L22:1772 weakly excites
  22. L23:391 weakly excites
  23. L23:765 weakly excites
  24. L21:2083 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.

1 values between -13.69 and -12.541 values between -12.54 and -11.381 values between -10.23 and -9.082 values between -9.08 and -7.935 values between -7.93 and -6.7710 values between -6.77 and -5.623 values between -5.62 and -4.479 values between -4.47 and -3.3233 values between -3.32 and -2.1699 values between -2.16 and -1.012390 values between -1.01 and 0.141302 values between 0.14 and 1.2925 values between 1.29 and 2.454 values between 2.45 and 3.605 values between 3.60 and 4.752 values between 4.75 and 5.903 values between 5.90 and 7.061 values between 7.06 and 8.211 values between 8.21 and 9.361 values between 9.36 and 10.512 values between 10.51 and 11.661 values between 11.66 and 12.822 values between 12.82 and 13.971 values between 13.97 and 15.121 values between 16.27 and 17.431 values between 18.58 and 19.731 values between 22.04 and 23.191 values between 31.25 and 32.41offnegativepositiveoften

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. SHIP
  3. SEE
  4. izzy
  5. }'",
  6. '},
  7. unda
  8. OLUMN
  9. ski
  10. чная

Suppresses

  1. can
  2. 可以
  3. 可以直接
  4. Can
  5. 可以在
  6. can
  7. _can
  8. Can
  9. →can
  10. 可以用