Neuron Explorer

Qwen3-0.6B

"you" as object inside a short clause

A neuron in layer 17 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 "you" when it sits inside a clause as the thing being acted on, for example "the tests you might have to check" or "whenever you can". It is weaker on words like "it" or "he" in the same position. The steering outputs did not clearly reflect this.

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

    can be cured or managed. To reduce your risk of heart and blood vessel problems, be sure to keep an eye on your cholesterol and blood pressure. The tests you might have to check your risk for heart and blood vessel problems depend on your age

  2. #2 peak

    non-biodegradable plastic, so recycle whenever you can. Cigarette butts are pollution and may be eaten by animals.

  3. #3 peak

    cotangent as⏎Both of these values can be decimals. The range of cotangent also has no restrictions; the ratios are any and all numbers,⏎The closer you get to the values where⏎the smaller the number on the bottom of your fraction

  4. #4 peak

    you must understand and apply some crucial National Electrical Code rules and regulation. Following these rules to the letter will not only assure you that your work will pass inspection, but that it will be a safe installation as

  5. #5 peak

    Trauma can lead to depression. Trauma can include things that happened directly to you or things that happened to a loved one. The unexpected loss of a loved one

  6. #6 peak

    In such cases, it becomes really important to use proper protective equipment in order to save yourself from an electric shock that may end up either injuring you too badly or worse.⏎Therefore, wearing shockproof gloves is a requirement by O

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] * \[ \text{[i]}\text{[i]}\text{[i]}\text{[i]}\text{[i]}\text{[i]}\text{[i]

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

    take a look at how animation on the Internet came to its present (and very cool) state. It pretty much all started with gifs…⏎.gif files are, it turns out, older than I am by about two years. They were introduced in

  2. #2 peak

    important part in the health of the tortoise. UVA & UVB help increase appetite, activity and induce reproductive behaviour. It also helps the tortoise produce it's own vitamin D3 which help's

  3. #3 peak

    their robots in competitions based on functionality.⏎Hopefully the students at the Ed Endres Boys & Girls Club will continue their interest in robotics and enter some tournaments, he continued.⏎“The kids have to practice and practice and work as a team to<|endoftext|>

  4. #4 peak

    that to U.S. health officials.⏎According to the Centers for Disease Control (CDC), the ALA's N-O-T program is the only smoking cessation program it rates as a "model program." In fact, the CDC does not list any

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. L16:1370 excites
  2. L10:293 excites
  3. L8:40 suppresses
  4. L14:382 excites
  5. L10:352 excites
  6. L4:1268 excites
  7. L6:549 suppresses
  8. L4:1855 excites
  9. L4:86 suppresses
  10. L4:136 excites
  11. L2:1653 suppresses
  12. L0:1167 excites
  13. L6:1922 excites
  14. L15:633 excites
  15. L4:1940 suppresses
  16. L9:2881 excites
  17. L7:67 excites
  18. L3:937 excites
  19. L3:1318 suppresses
  20. L16:703 suppresses
  21. L1:1549 suppresses
  22. L4:2453 weakly suppresses
  23. L14:1997 weakly excites
  24. L3:2210 weakly suppresses

Feeds later layers

  1. L18:770 strongly excites
  2. L18:1043 strongly excites
  3. L18:2165 excites
  4. L18:582 suppresses
  5. L20:1833 suppresses
  6. L18:918 excites
  7. L19:1779 suppresses
  8. L21:1276 suppresses
  9. L21:2699 excites
  10. L19:2483 suppresses
  11. L18:2167 excites
  12. L18:2575 excites
  13. L21:1232 weakly suppresses
  14. L20:1446 weakly excites
  15. L18:1459 weakly excites
  16. L22:1346 weakly suppresses
  17. L27:1047 weakly suppresses
  18. L23:1521 weakly excites
  19. L18:687 weakly suppresses
  20. L19:149 weakly suppresses
  21. L22:545 weakly excites
  22. L21:1040 weakly suppresses
  23. L22:2602 weakly suppresses
  24. L24:67 weakly 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.

483 values between -1.69 and -0.992552 values between -0.99 and -0.29736 values between -0.29 and 0.4216 values between 0.42 and 1.1214 values between 1.12 and 1.8311 values between 1.83 and 2.536 values between 2.53 and 3.239 values between 3.23 and 3.937 values between 3.93 and 4.644 values between 4.64 and 5.348 values between 5.34 and 6.043 values between 6.04 and 6.754 values between 6.75 and 7.454 values between 7.45 and 8.154 values between 8.15 and 8.863 values between 8.86 and 9.562 values between 9.56 and 10.261 values between 10.26 and 10.974 values between 10.97 and 11.673 values between 11.67 and 12.372 values between 12.37 and 13.085 values between 13.78 and 14.485 values between 14.48 and 15.192 values between 15.89 and 16.591 values between 16.59 and 17.301 values between 17.30 and 18.003 values between 18.00 and 18.704 values between 18.70 and 19.411 values between 19.41 and 20.113 values between 20.11 and 20.811 values between 20.81 and 21.523 values between 22.92 and 23.632 values between 23.63 and 24.331 values between 25.73 and 26.44offnegativepositiveoften

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. バー
  2. scape
  3. mates
  4. chants
  5. ENCY
  6. rawid
  7. stance
  8. oned
  9. メント
  10. วด

Suppresses

  1. 其所
  2. 其
  3. 他还
  4. 他又
  5. xaf
  6. its
  7. 他
  8. 臾
  9. 他表示
  10. their