Neuron Explorer

Qwen3-0.6B

factor nouns like education in lists

A neuron in layer 12 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

This neuron fires on words naming a factor that shapes an outcome, such as 'education' or 'income' in a list of influences like 'age, education, and income'. Turning it up did not make the model write about those factors, so the link is only in the text it reads.

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

    weekly earnings by worker characteristics, including educational attainment on the right-hand side:⏎Source: New York Times, Bureau of Labor Statistics

  2. #2 peak

    realized by means of a dedicated MV substation usually designed "Main substation". Depending on its size and specific criteria mainly related to the loads (Rated voltage, number, power, location, etc…), the installation may include additional substations designed

  3. #3 peak

    natural gas are organic substances found in⏎solid, liquid and gaseous forms respectively.⏎Physical Properties of Minerals⏎Minerals can be identified by their hardness,⏎colour, the way they reflect light (lustre), the⏎way they break (

  4. #4 peak

    are two distinct numbers regarding the pay gap: unadjusted versus adjusted pay gap. The latter takes into account factors that contribute to the gap, such as differences in education, job experience, hours worked and type of occupation.⏎UN Women Georgia,

  5. #5 peak

    2019.<|endoftext|>Social medicine is the study of health in its widest sense. It recognizes the broad determinants of health – income and poverty, education, environmental factors such as housing and transport – as well as health care and genetic

  6. #6 peak

    likely to serve you for a very long duration.⏎Various things should be put into consideration before purchasing a propeller. The person buying the propeller must consider the size of the boat first to determine the size of propeller they need for their boat

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: (this is a problem, if any) (this is a problem, if any) (this is a problem, if any) (this is a problem, if any) (this is a problem, if any) (this is

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

    safety is important year-round, but becomes especially important during the holidays when the warm, cozy glow of candles is a tradition for many,

  2. #2 peak

    Medical history questions may include:⏎- When did the excessive yawning begin?⏎- How many times do you yawn per hour or day?⏎- Is it worse in the morning, after lunch, or during exercise?⏎- Is it worse in certain

  3. #3 peak

    made from ingredients such as meat, grains, vegetables, and fruits. Dog food may also contain vitamins, minerals, and other supplements. Some dog foods are for specific types of dogs such as puppies or senior dogs. There are many different types of

  4. #4 peak

    may have higher exposures to mould.⏎Is mould harmful to my health?⏎Exposure to mould does not always have negative effects. However, the following individuals may be more sensitive to mould:⏎- pregnant women⏎- the elderly⏎- those with asthma

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. L9:1505 excites
  2. L10:982 suppresses
  3. L11:1380 suppresses
  4. L8:2195 weakly excites
  5. L11:1744 weakly excites
  6. L10:495 weakly excites
  7. L11:1688 weakly excites
  8. L10:2111 weakly suppresses
  9. L11:2246 weakly excites
  10. L11:1793 weakly suppresses
  11. L11:1310 weakly suppresses
  12. L10:1400 weakly excites
  13. L11:316 weakly suppresses
  14. L11:613 weakly suppresses
  15. L8:2233 weakly excites
  16. L10:534 weakly excites
  17. L9:1950 weakly excites
  18. L11:2664 weakly excites
  19. L10:1570 weakly suppresses
  20. L9:377 weakly suppresses
  21. L7:1846 weakly suppresses
  22. L8:2690 weakly suppresses
  23. L9:217 weakly suppresses
  24. L7:1897 weakly excites

Feeds later layers

  1. L23:720 suppresses
  2. L13:2435 suppresses
  3. L14:406 suppresses
  4. L14:1936 suppresses
  5. L16:1729 suppresses
  6. L14:1664 suppresses
  7. L14:1164 excites
  8. L14:253 excites
  9. L14:2211 suppresses
  10. L14:2333 weakly suppresses
  11. L15:2154 weakly suppresses
  12. L17:1970 weakly excites
  13. L22:2137 weakly suppresses
  14. L14:445 weakly suppresses
  15. L14:1792 weakly suppresses
  16. L14:2695 weakly suppresses
  17. L16:919 weakly excites
  18. L14:640 weakly suppresses
  19. L15:1952 weakly suppresses
  20. L13:171 weakly suppresses
  21. L14:2910 weakly excites
  22. L13:733 weakly suppresses
  23. L19:1063 weakly suppresses
  24. L25:2352 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 -2.50 and -2.401 values between -2.29 and -2.192 values between -1.88 and -1.781 values between -1.78 and -1.673 values between -1.57 and -1.472 values between -1.47 and -1.365 values between -1.36 and -1.263 values between -1.26 and -1.161 values between -1.16 and -1.056 values between -1.05 and -0.954 values between -0.95 and -0.8510 values between -0.85 and -0.749 values between -0.74 and -0.647 values between -0.64 and -0.5430 values between -0.54 and -0.4333 values between -0.43 and -0.3397 values between -0.33 and -0.23262 values between -0.23 and -0.12990 values between -0.12 and -0.021875 values between -0.02 and 0.08338 values between 0.08 and 0.19132 values between 0.19 and 0.2948 values between 0.29 and 0.3921 values between 0.39 and 0.509 values between 0.50 and 0.608 values between 0.60 and 0.701 values between 0.70 and 0.812 values between 0.81 and 0.911 values between 0.91 and 1.011 values between 1.01 and 1.123 values between 1.12 and 1.222 values between 1.53 and 1.63offnegativepositiveoften

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. เซล
  2. Equivalent
  3. otre
  4. 的本质
  5. equivalent
  6. กระท
  7. lu
  8. cop
  9. lice
  10. 头脑

Suppresses

  1. 特定
  2. _specific
  3. special
  4. 特别
  5. เฉพาะ
  6. particular
  7. 特殊
  8. pecific
  9. specific
  10. específ