Neuron Explorer

Qwen3-0.6B

"the" in "on the other hand"

A neuron in layer 21 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 lights up on the word "the" in the phrase "on the other hand", the kind used to set up a contrast between two ideas. Turning it up made the model's writing drift into loops and odd phrasing, not contrast sentences, so the steering results don't clearly show what it does.

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

    Hard work can lead to a deep sense of accomplishment and mastery, but can also be time-consuming and may not be the most effective approach. Smart work, on the other hand, can save time and increase productivity, but may lead to a lack

  2. #2 peak

    to the environment, due to the emissions that come as a result of its use.⏎The Good News⏎Vaperma’s membrane-based Siftek, on the other hand, is expected to reduce emissions significantly. And in addition to this,

  3. #3 peak

    are poor, with 23 percent living “in misery, on one meal a day and not a very nourishing meal at that.” The rich, on the other hand, are very rich indeed. “There is no inheritance tax, and

  4. #4 peak

    pure glucose as the comparison point.⏎Foods with a high GI, such as white bread and white rice, cause a rapid increase in blood sugar levels. On the other side, foods with low GI, such as whole grains and sweet potato,

  5. #5 peak

    complex problems at speeds that were previously unimaginable. Traditional computers rely on bits, which can represent either a 0 or a 1. Quantum computers, on the other hand, use quantum bits or qubits, which can exist in multiple<|endoftext|>

  6. #6 peak

    as patents and trademarks. These assets are categorized as current or non-current, depending on their liquidity and expected conversion into cash within a year.⏎Liabilities, on the other hand, encompass the company's financial obligations and debts. These can include loans

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 middle between the top and the above the top (the top above the top above the top... and above the top above the top above the top above the top... between the top and the above the top... above the top... above

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

    this portion of the course, via discussions, you will be asked to critically think and debate varying interpretations of the empirical evidence in relation to the science being brought to your attention<|endoftext|>Economics– Important Questions Bank for Rajasthan Board HSC 20

  2. #2 peak

    partition alignment is something new to me. I have heard of it before but I have never used nor understood it’s meaning until a colleague at work brought it to my attention for a customer we are trying to help out.⏎The information in this post

  3. #3 peak

    of Suleiman the Magnificent. However, after continued unrest in the interior, the Turks evacuated the region in 1630, leaving it in the hands of the Shiite Zaidi imams of Sanaa. Muhammad Ali

  4. #4 peak

    00 tons per year.⏎Reusable, Erasable Paper⏎Even with all the digital technology we have at our fingertips, paper is one of those things we seem to inevitably waste. It seems

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. L18:1375 strongly suppresses
  2. L18:483 strongly excites
  3. L18:2356 strongly excites
  4. L19:2820 strongly suppresses
  5. L17:1434 excites
  6. L10:262 suppresses
  7. L19:939 suppresses
  8. L16:1807 suppresses 'in' after a motion verb
  9. L19:1581 suppresses
  10. L10:1603 excites
  11. L18:2677 excites
  12. L16:2120 suppresses
  13. L4:1966 suppresses
  14. L15:853 suppresses
  15. L5:490 excites
  16. L14:1389 suppresses
  17. L14:1848 excites
  18. L20:2201 excites 'the' in location phrases like 'at the end'
  19. L18:866 suppresses
  20. L3:311 suppresses
  21. L20:2873 suppresses
  22. L17:646 suppresses
  23. L17:2147 excites
  24. L17:303 suppresses

Feeds later layers

  1. L27:1363 strongly suppresses
  2. L26:814 strongly suppresses
  3. L23:1437 suppresses
  4. L25:1921 suppresses
  5. L25:1107 suppresses
  6. L25:925 suppresses
  7. L23:476 suppresses
  8. L26:585 suppresses
  9. L22:1026 suppresses
  10. L25:1660 suppresses
  11. L23:2829 suppresses
  12. L25:1865 suppresses 'in' in 'keep in mind'
  13. L22:938 suppresses
  14. L24:1796 suppresses
  15. L22:1110 suppresses
  16. L26:1080 suppresses
  17. L23:512 suppresses
  18. L22:2577 suppresses
  19. L27:906 suppresses
  20. L22:885 suppresses
  21. L23:353 weakly suppresses
  22. L26:1662 weakly suppresses
  23. L27:1179 weakly suppresses
  24. L27:1269 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 -11.93 and -11.421 values between -10.41 and -9.901 values between -9.90 and -9.392 values between -9.39 and -8.881 values between -8.88 and -8.381 values between -8.38 and -7.871 values between -7.87 and -7.363 values between -7.36 and -6.853 values between -6.35 and -5.849 values between -5.84 and -5.333 values between -5.33 and -4.825 values between -4.82 and -4.327 values between -4.32 and -3.817 values between -3.81 and -3.3010 values between -3.30 and -2.7916 values between -2.79 and -2.2924 values between -2.29 and -1.7829 values between -1.78 and -1.2737 values between -1.27 and -0.7676 values between -0.76 and -0.251851 values between -0.25 and 0.251587 values between 0.25 and 0.76201 values between 0.76 and 1.2716 values between 1.27 and 1.782 values between 1.78 and 2.285 values between 2.28 and 2.792 values between 2.79 and 3.302 values between 3.30 and 3.812 values between 3.81 and 4.311 values between 4.31 and 4.821 values between 5.84 and 6.351 values between 7.87 and 8.38offnegativepositiveoften

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. une
  2. aturing
  3. UNE
  4. ework
  5. 팅
  6. -Man
  7. ă
  8. sorts
  9. lasses
  10. сложн

Suppresses

  1. 身上
  2. 里面
  3. 面前
  4. 那里
  5. 之中
  6. 裡面
  7. 活动中
  8. 环境中
  9. 其中
  10. 人群中