Neuron Explorer

Qwen3-0.6B

"the" in size comparisons

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

The neuron fires on the word "the" in phrases that compare something's size to a familiar object, like "about the size of a small dog." Turned up, the model started repeating "the temperature of the temperature", so the link to the trigger is loose.

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

    cken<|endoftext|>Ice shelf the size of Connecticut at risk of collapsing⏎- Grades: 3–5, 6–8⏎A hunk of ice about the size of the state of Connecticut has started to collapse in Antarctica. According to satellite

  2. #2 peak

    is a changing mole.⏎What does Stage 1 melanoma look like?⏎Stage I melanoma is no more than 1.0 millimeter thick (about the size of a sharpened pencil point), with or

  3. #3 peak

    What Are Bed Bugs?⏎Before we dive into the many different bed bug types, let’s talk some about what bed bugs are. Bed bugs are small insects about the size, shape, and color of an apple seed. These insects live in fabric

  4. #4 peak

    of margarine is the size of one dice⏎- 3 ounces of meat is the size of a deck of cards⏎- 1 cup of pasta is the size of a baseball⏎- 1½ ounces of cheese is the size of

  5. #5 peak

    for students without prior assembly-language background to master the many facets of pointer usage.<|endoftext|>International Day for the Remembrance of the Slave Trade and its Abolition⏎On the night of

  6. #6 peak

    — the atomic jitters that pass through our world when two black holes bash together in a distant galaxy are on the order of a billionth of a billionth the diameter of an atom. LIGO detects them by proxy, using high powered lasers

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: I need to a certain project that I need to a project that I need to a project that I need to a project that I need to a project that I need to a project that I need to a project that I need to a project

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

    .⏎The scientists wanted to examine if these observed effects were reversible. They returned mice to their normal diet after four weeks and were surprised to find that blood flow and blood

  2. #2 peak

    been working on important experiments with metal alloys for aircraft. He had the results of seven years' labour in a portfolio with him and rather than let the research fall into German hands, (Poland having been invaded by Germany), he has hidden it

  3. #3 peak

    how much further we need to go," said Dereth Glance, US Commissioner.⏎"This report is only the beginning as the Commission is defining important keystone indicators of ecological and human health that can be used over the long-term to assess

  4. #4 peak

    teacher gets feedback about which questions were most attempted by students, which questions most students got wrong, etc. This helps teachers and school management to get an overall picture of what the real state of students is. Also, this provides critical feedback to teachers

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. L20:2692 excites
  2. L19:1357 suppresses
  3. L19:2385 excites
  4. L19:3054 suppresses
  5. L19:1742 excites
  6. L20:712 excites
  7. L18:1918 excites
  8. L19:1403 suppresses
  9. L19:566 suppresses
  10. L20:1342 excites
  11. L17:1913 excites
  12. L18:2718 excites
  13. L18:2809 suppresses
  14. L19:1080 excites
  15. L18:2344 excites
  16. L20:1846 excites
  17. L20:2587 suppresses
  18. L19:1808 suppresses
  19. L17:2443 excites
  20. L19:3050 suppresses
  21. L18:748 suppresses
  22. L20:47 suppresses
  23. L17:1207 suppresses
  24. L20:1417 excites

Feeds later layers

  1. L27:127 suppresses
  2. L27:399 excites
  3. L26:507 excites
  4. L23:94 suppresses
  5. L23:1486 excites
  6. L26:1287 excites
  7. L24:49 weakly suppresses
  8. L22:241 weakly suppresses
  9. L22:4 weakly excites
  10. L27:666 weakly suppresses
  11. L25:1796 weakly excites
  12. L24:2538 weakly suppresses
  13. L27:1038 weakly suppresses
  14. L23:24 weakly excites
  15. L25:2497 weakly excites
  16. L22:171 weakly excites
  17. L22:1620 weakly excites "the" after abstract verbs like allows for
  18. L24:1366 weakly suppresses
  19. L25:589 weakly suppresses
  20. L26:528 weakly excites
  21. L26:1816 weakly suppresses
  22. L23:672 weakly suppresses
  23. L26:363 weakly suppresses
  24. L25:525 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 -5.84 and -5.441 values between -5.44 and -5.043 values between -4.24 and -3.842 values between -3.04 and -2.643 values between -2.64 and -2.245 values between -2.24 and -1.8413 values between -1.84 and -1.4531 values between -1.45 and -1.0548 values between -1.05 and -0.65373 values between -0.65 and -0.252094 values between -0.25 and 0.151020 values between 0.15 and 0.55167 values between 0.55 and 0.9543 values between 0.95 and 1.3524 values between 1.35 and 1.7516 values between 1.75 and 2.1513 values between 2.15 and 2.5515 values between 2.55 and 2.954 values between 2.95 and 3.358 values between 3.35 and 3.758 values between 3.75 and 4.155 values between 4.15 and 4.543 values between 4.54 and 4.941 values between 4.94 and 5.342 values between 5.34 and 5.742 values between 5.74 and 6.141 values between 6.14 and 6.541 values between 6.94 and 7.341 values between 9.73 and 10.13offnegativepositiveoften

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. The
  2. _the
  3. the
  4. the
  5. _THE
  6. THE
  7. The
  8. .the
  9. →the
  10. -the

Suppresses

  1. arian
  2. odian
  3. icle
  4. indrical
  5. ingleton
  6. 亿元以上
  7. ard
  8. film
  9. atom
  10. illian