Neuron Explorer

Qwen3-0.6B

the word 'hidden' or concealed

A neuron in layer 20 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron lights up on the word 'hidden' and close synonyms like 'lurking' and 'implicit', meaning something concealed. When turned up, the model sometimes brought in the word 'hidden' itself, as in a story called The Hidden Unconscious.

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

    to practice more often). All this before they ever solve the math problems. Best of all, most of my students really enjoyed this activity that was packed with (hidden) learning.⏎Here's how the completed assignment looks:⏎Six little math problems,

  2. #2 peak

    the stack of languages like C or garbage collected in languages like Java.⏎Byers, Simon. Scalable Exploitation of, and Responses to Information Leakage Through Hidden Data in Published Documents, AT&T Research, April 2003⏎

  3. #3 peak

    Back in the early 20th century, the mountains and hollers of Polk County were infamous for hosting moonshiners and bootleggers, hidden away from the long arm of the government “revenooers.” But for

  4. #4 peak

    , post-tensioned concrete, concrete in contact with chlorides, parking structures, piers, bridges and marine concrete.<|endoftext|>Blended to oblivion, hidden in a stew, mashed to a paste, however you get your

  5. #5 peak

    Aesopean or other proverbial gem of wisdom. Recognizing those words in the gaps left between the stroked terms can help spot that last few stubborn hidden letters.⏎The grid can be resized, flipped and rotated in all directions. This

  6. #6 peak

    the prize, the one has been around the longest, that we use more often and has more impact on our health, wellbeing and comfort, is lurking in the corner of the bathroom - the lavatory. A device that is so

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: A very recent, very young, and very much less than 100 years old (less than 100 years) of a **v, v, v: 10: 10: 10:

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

    product information, but the⏎site also needs to let engineers access the information they need⏎in as few clicks as possible. The one component engineers have in⏎short supply is time,” shares Kevin Hess, Mouser

  2. #2 peak

    brain fog is an ongoing issue for you, taking a look at and adjusting your lifestyle habits might be warranted. Let’s look at some common causes:⏎In the short-term, stress can make you irritable, anxious, distracted and forget<|endoftext|>R

  3. #3 peak

    hardly least, we have precious few black figures whom we feel good about eulogizing these days, people who mainstream America would define as a uniter in short, those who have dropped the cause of black freedom or softened it sufficiently to merit

  4. #4 peak

    until the crickets fall asleep; then they are frozen. "You still have to kill an animal, but it’s more humane," he says.⏎In the short-term, Dzamba imagines that the contraptions could find a market

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. L16:2784 suppresses
  2. L12:1343 suppresses
  3. L10:2262 excites
  4. L17:1621 weakly suppresses
  5. L7:334 weakly suppresses
  6. L19:884 weakly excites
  7. L14:2214 weakly excites
  8. L12:1564 weakly suppresses
  9. L7:1741 weakly excites
  10. L10:2523 weakly suppresses
  11. L12:2992 weakly suppresses
  12. L3:1087 weakly excites
  13. L14:2628 weakly excites
  14. L12:2331 weakly excites
  15. L4:2039 weakly suppresses
  16. L10:1989 weakly suppresses
  17. L16:2688 weakly suppresses
  18. L0:1946 weakly suppresses
  19. L18:196 weakly excites
  20. L13:2638 weakly excites
  21. L18:2394 weakly suppresses
  22. L19:2802 weakly excites
  23. L15:377 weakly excites "even" in concessive phrases
  24. L15:2494 weakly excites

Feeds later layers

  1. L22:1701 excites
  2. L24:1815 excites
  3. L25:2502 excites
  4. L27:2977 excites
  5. L25:1834 weakly excites
  6. L25:1708 weakly excites
  7. L24:1915 weakly suppresses 'in' after a noun phrase
  8. L23:2713 weakly suppresses
  9. L22:1294 weakly suppresses
  10. L22:3001 weakly excites
  11. L21:2676 weakly excites
  12. L21:2137 weakly excites
  13. L22:2528 weakly excites
  14. L22:1293 weakly suppresses
  15. L21:1648 weakly suppresses
  16. L25:2285 weakly excites
  17. L23:990 weakly suppresses
  18. L25:1562 weakly excites
  19. L22:2208 weakly suppresses
  20. L22:831 weakly excites
  21. L21:1051 weakly excites
  22. L24:796 weakly excites
  23. L25:1352 weakly suppresses
  24. L21:2116 weakly excites third digit of years and numbers

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 -6.94 and -6.272 values between -4.94 and -4.273 values between -4.27 and -3.605 values between -3.60 and -2.937 values between -2.93 and -2.2714 values between -2.27 and -1.6043 values between -1.60 and -0.93691 values between -0.93 and -0.272734 values between -0.27 and 0.40295 values between 0.40 and 1.0759 values between 1.07 and 1.7421 values between 1.74 and 2.4011 values between 2.40 and 3.076 values between 3.07 and 3.741 values between 3.74 and 4.413 values between 4.41 and 5.076 values between 5.07 and 5.743 values between 5.74 and 6.411 values between 7.07 and 7.741 values between 11.08 and 11.741 values between 19.08 and 19.75offnegativepositiveoften

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. 隐
  2. 暗
  3. 隐藏
  4. hidden
  5. 潜
  6. Hidden
  7. 隐形
  8. 悄悄
  9. 隱
  10. implicit

Suppresses

  1. itz
  2. bons
  3. ycin
  4. ortion
  5. orough
  6. 柁
  7. зв
  8. فيل
  9. die
  10. 宁县