Neuron Explorer

Qwen3-0.6B

subjects of conditional or advice clauses

A neuron in layer 19 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

It fires on the word that introduces who or what a conditional or possible situation is about, such as 'you' in 'or you may require' or 'they' in 'they can arrange'. Turning it up pushed the text toward hypothetical scenarios, and turning it down pushed it toward 'recent studies' phrasing.

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

    be. For example, you may be happy with a confidence level of 95 percent (giving a margin error of 0.05), or you may require a tighter accuracy of a 98 percent confidence level (a margin

  2. #2 peak

    it is high probability that the hypothesis has been refuted.⏎If the hypothesis failed in their experiments, repeatedly, the scientist can do one of two things:⏎- He

  3. #3 peak

    a little and sometimes nothing at all. For example, we may know that a population is normally distributed (e.g., for heights, weights or IQs), we may know that there is a bimodal distribution (as often happens with class

  4. #4 peak

    The period during which a cell or battery is capable of operating above a specified capacity or efficiency performance level. Life may be measured in cycles and/or years, depending on the type of service for which the cell or battery is intended.⏎Btu (

  5. #5 peak

    fine particles⏎- Mix detergent with warm water, according to the instructions on the label⏎- Use a sponge or rag to apply the detergent

  6. #6 peak

    provides critical feedback to teachers about how well students are performing in the concepts taught by them. If teachers find that most students are going wrong in a particular concept, they can arrange for remedial classes for such concepts and help students gain proficiency in the

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 current version of the original $A$ and $B$ do not have an exact mathematical proof, but have been noted recently in the article [1] that a recent mathematical paper by the authors of the 2016 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

    as Rheumatoid Arthritis (RA). We’ve seen great medical advances in the treatment of and even preventing certain diseases once we’re able to predict that a person is prone to getting the disease.⏎It just may be that rheumatoid arthritis

  2. #2 peak

    need to be naturalized if born in the U.S.⏎- Census The 1900 and 1910 censuses ask if a person is⏎naturalized and 1920 further asks the year of natural

  3. #3 peak

    tested. Talking to their health care providers may help parents identify the option that is right for them.⏎All U.S. states have neonatal screening for CA

  4. #4 peak

    any of the components listed above are not belt driven, you must:⏎- Tell the examiner which component(s) are not belt driven.⏎- Make sure component(s) are operating properly, are not damaged or leaking, and are mounted securely.⏎-

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. L18:1338 suppresses
  2. L17:1611 suppresses
  3. L13:1506 excites
  4. L16:837 excites
  5. L15:1879 excites
  6. L16:730 excites
  7. L15:742 excites
  8. L16:835 suppresses
  9. L18:2656 suppresses
  10. L15:878 suppresses
  11. L15:782 excites
  12. L17:1945 suppresses
  13. L17:951 excites
  14. L14:1477 excites
  15. L15:607 suppresses
  16. L15:1641 excites
  17. L16:3069 suppresses
  18. L17:788 suppresses
  19. L16:2332 suppresses
  20. L18:1038 suppresses
  21. L12:611 excites
  22. L12:717 excites
  23. L13:1153 excites
  24. L15:1623 excites

Feeds later layers

  1. L20:589 suppresses
  2. L20:388 excites
  3. L21:1093 suppresses
  4. L21:1461 suppresses
  5. L20:2973 excites
  6. L20:2318 suppresses 'recent' in time phrases
  7. L20:1396 suppresses
  8. L20:2894 suppresses
  9. L23:1173 excites
  10. L20:2110 suppresses
  11. L20:2939 suppresses
  12. L20:1666 suppresses
  13. L24:499 suppresses
  14. L20:726 suppresses
  15. L21:2121 excites
  16. L21:1034 suppresses
  17. L21:2506 suppresses
  18. L23:1982 suppresses
  19. L21:2592 suppresses
  20. L21:169 excites
  21. L20:990 suppresses
  22. L21:1729 suppresses
  23. L22:2268 suppresses empty slot before a number
  24. L22:1358 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 -4.72 and -4.481 values between -4.24 and -4.014 values between -4.01 and -3.771 values between -3.77 and -3.531 values between -3.53 and -3.294 values between -3.29 and -3.053 values between -3.05 and -2.811 values between -2.81 and -2.575 values between -2.57 and -2.337 values between -2.33 and -2.0910 values between -2.09 and -1.857 values between -1.85 and -1.6110 values between -1.61 and -1.3725 values between -1.37 and -1.1353 values between -1.13 and -0.90143 values between -0.90 and -0.66354 values between -0.66 and -0.42589 values between -0.42 and -0.181075 values between -0.18 and 0.06544 values between 0.06 and 0.30328 values between 0.30 and 0.54229 values between 0.54 and 0.78145 values between 0.78 and 1.02113 values between 1.02 and 1.2663 values between 1.26 and 1.5043 values between 1.50 and 1.7438 values between 1.74 and 1.9832 values between 1.98 and 2.2227 values between 2.22 and 2.4611 values between 2.46 and 2.6910 values between 2.69 and 2.9310 values between 2.93 and 3.179 values between 3.17 and 3.414 values between 3.41 and 3.652 values between 3.65 and 3.892 values between 3.89 and 4.131 values between 4.37 and 4.611 values between 4.61 and 4.85offnegativepositiveoften

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. 会
  2. owning
  3. omm
  4. sume
  5. 있을
  6. 就会
  7. Hanna
  8. 会长
  9. 会有
  10. 一定会

Suppresses

  1. recent
  2. lately
  3. recently
  4. 至今
  5. 近年来
  6. recent
  7. 近年
  8. 著名的
  9. 到现在
  10. amous