Neuron Explorer

Qwen3-0.6B

blank spot where an age or number goes

A neuron in layer 12 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

Fires on the spot just before a number, especially after words like "age", "between" or "older than", so the model is expecting a figure next. Turning it up made the model drift into talk about numbers and percentages, which fits this pattern loosely.

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

    ethnic minority students. Samples: One hundred and five preservice teachers (90 female) with a mean age of 26.20 years (teaching experience: 57.5

  2. #2 peak

    In 2018, about 21.2 million people age 12 and older, or approximately 1 in 13 people in this

  3. #3 peak

    had almost 150 years to adapt to a change from 10% to 20% in the proportion of the population that was older than 60 years, places like Brazil, China and India will have slightly more than

  4. #4 peak

    children who didn’t have the disease. Each group shared the same gender, age, and combination of genetic risk factors for celiac disease. The children were between 15 months and 8 years old when they were diagnosed for celiac disease

  5. #5 peak

    The history of art can be traced back to cave paintings of about 15000 BC. Gotell, E., Brown, S.,

  6. #6 peak

    of health, how it manifests and the consequences for people and society.<|endoftext|>Fact 1: The world’s population is rapidly ageing⏎The number of people aged 60 years or older will rise from 900 million to 2

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 role of a teacher in the early years. The first thing is that they can't have any problem with this or anything else. The first thing is to get the students into the early years. The age of teaching. The age of

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

    1619 and set an example for others to follow. As Virginians migrated in search of new opportunity, they took these lessons of self-rule they had learned and put them

  2. #2 peak

    College in Oakdale, NY, who also teaches elementary school special education classes. When reading to her students, Granieri thinks out loud — commenting on how the text may add to the child’s knowledge of the topic. For example

  3. #3 peak

    pursuits, whether you are an amateur or professional.⏎Perfectionism is unproductive, however, when it is linked to excessively high standards and is driven by a fear of failure.⏎Back to top⏎Types of Perfectionism⏎Perfectionism

  4. #4 peak

    lescent anorexia develops in boys too.”⏎Anorexia Nervosa Symptoms⏎A person with anorexia usually displays these:⏎- Intense fear of gaining weight or getting overweight even when they could be underweight.⏎- Would have

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. L11:298 weakly excites
  2. L11:1284 weakly excites
  3. L8:756 weakly excites
  4. L6:1140 weakly excites
  5. L7:2395 weakly excites
  6. L10:1295 weakly suppresses
  7. L2:1906 weakly excites
  8. L11:44 weakly suppresses
  9. L8:2309 weakly excites
  10. L9:837 weakly suppresses
  11. L10:844 weakly excites
  12. L5:279 weakly suppresses
  13. L2:94 weakly excites
  14. L11:2035 weakly suppresses
  15. L7:1929 weakly excites
  16. L8:64 weakly excites
  17. L10:2768 weakly excites
  18. L7:2889 weakly excites
  19. L5:2065 weakly excites
  20. L7:469 weakly suppresses
  21. L2:1172 weakly suppresses
  22. L8:1062 weakly excites
  23. L6:1776 weakly suppresses street and urban place names
  24. L0:3050 weakly excites

Feeds later layers

  1. L19:2939 weakly suppresses
  2. L13:2527 weakly suppresses
  3. L15:1860 weakly excites
  4. L21:1231 weakly excites
  5. L14:2708 weakly excites
  6. L15:233 weakly suppresses
  7. L22:359 weakly excites
  8. L17:1967 weakly suppresses
  9. L18:523 weakly suppresses
  10. L13:1171 weakly excites
  11. L13:2466 weakly excites
  12. L22:1755 weakly excites
  13. L13:2414 weakly suppresses
  14. L13:2362 weakly excites
  15. L18:2180 weakly excites
  16. L14:1647 weakly suppresses
  17. L20:1283 weakly excites
  18. L13:3027 weakly excites
  19. L23:294 weakly suppresses
  20. L23:2299 weakly excites
  21. L13:314 weakly excites
  22. L18:991 weakly excites
  23. L21:852 weakly excites
  24. L15:1815 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.

1 values between -3.88 and -3.712 values between -3.71 and -3.551 values between -3.05 and -2.881 values between -2.55 and -2.383 values between -2.22 and -2.054 values between -2.05 and -1.892 values between -1.89 and -1.726 values between -1.72 and -1.5514 values between -1.55 and -1.3920 values between -1.39 and -1.2228 values between -1.22 and -1.0643 values between -1.06 and -0.8969 values between -0.89 and -0.73125 values between -0.73 and -0.56196 values between -0.56 and -0.39328 values between -0.39 and -0.23630 values between -0.23 and -0.06906 values between -0.06 and 0.11659 values between 0.11 and 0.27401 values between 0.27 and 0.44197 values between 0.44 and 0.60125 values between 0.60 and 0.7772 values between 0.77 and 0.9428 values between 0.94 and 1.1021 values between 1.10 and 1.2711 values between 1.27 and 1.433 values between 1.43 and 1.603 values between 1.60 and 1.764 values between 1.76 and 1.931 values between 1.93 and 2.103 values between 2.10 and 2.261 values between 2.59 and 2.76offnegativepositiveoften

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. apest
  2. 就被
  3. abe
  4. setDefault
  5. 专项行动
  6. aspers
  7. >{@
  8. atti
  9. activity
  10. 地址

Suppresses

  1. urple
  2. ieve
  3. 匦
  4. ерь
  5. ющую
  6. unya
  7. ikk
  8. 明日
  9. �
  10. ivel