Neuron Explorer

Qwen3-0.6B

"according" after a comma in news

A neuron in layer 2 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

This neuron lights up on the word "according" when it follows a comma and a claim, as in "..., according to the study". This is the usual way news articles credit a source. Turning it up did not clearly change the model's writing in a related way.

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

    disease that poses a major health risk for 28 million Americans, according to the National Osteoporosis Foundation. Vashishth and researchers at

  2. #2 peak

    daily functioning. This condition, known as mild cognitive impairment, progresses to Alzheimer’s in roughly 10% to 15% of cases each year, according to the study.⏎Forty-five percent of the participants went on to develop serious

  3. #3 peak

    2.5 billion people do not have a formal account at a financial institution, according to the World Bank’s Global Financial Inclusion Database. As a result, most

  4. #4 peak

    be a safe and effective way to relieve chronic pain in adolescents suffering from a variety of chronic pain conditions, according to a report published by Pain Medicine News.⏎The purpose of the study was to

  5. #5 peak

    someone like me<|endoftext|>By Sarah Zagorski⏎In the United Kingdom, women in their early thirties are now more likely to have abortions than teenagers, according to a new report from the Department of Health. These women are usually unmarried and

  6. #6 peak

    number of Americans who have a history of cancer was 11.7 million in 2007, or nearly 4% of the population, according to new figures from the Centers for Disease Control and Prevention (CDC).⏎That's

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 am a little guy, I have a lot of things to do, I am the one, I have a lot of things to do, I have a lot of things to do, I have a lot of things to do,

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

    human genome has about 23,000 genes over 3Gb of DNA). Genes were found corresponding to genes for energy metabolism, both aerobic (used by the juveniles) and anaerobic (used by the adults).

  2. #2 peak

    cooperative learning). So the Information seemed to be regarded as permeable and not owned by the teacher. From teachers' perspective, it was necessary to employ both traditional (e.g., lecturing) and non-traditional (e.g., encouraging

  3. #3 peak

    015]⏎Conway's Game of Life⏎Conway's Game of Life consists of a world of cells on a grid. These cells are either alive (on) or dead (off). The state of cell is determined by a set

  4. #4 peak

    ighbone and one medial meniscus under the inner knob of the thighbone. Each meniscus acts as a natural cushion between the thighbone (femur) and shinbone (tibia). The two cushions prevent excess

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. L1:989 weakly excites
  2. L0:1040 weakly excites
  3. L1:602 weakly excites
  4. L1:951 weakly excites
  5. L0:868 weakly suppresses
  6. L0:685 weakly suppresses
  7. L0:902 weakly suppresses
  8. L1:282 weakly suppresses
  9. L1:1055 weakly suppresses
  10. L1:1246 weakly excites "word" after a language origin phrase
  11. L0:2929 weakly suppresses
  12. L0:1489 weakly suppresses
  13. L1:1757 weakly suppresses
  14. L0:479 weakly excites
  15. L0:1337 weakly suppresses
  16. L1:604 weakly suppresses
  17. L1:1689 weakly excites
  18. L0:972 weakly suppresses
  19. L1:1140 weakly suppresses
  20. L1:898 weakly suppresses
  21. L0:406 weakly suppresses
  22. L1:323 weakly excites
  23. L1:1373 weakly suppresses
  24. L1:1036 weakly suppresses

Feeds later layers

  1. L12:878 suppresses
  2. L15:232 weakly suppresses
  3. L6:990 weakly suppresses
  4. L15:191 weakly excites
  5. L9:2713 weakly excites
  6. L27:1356 weakly suppresses
  7. L20:2116 weakly excites
  8. L18:1563 weakly excites
  9. L10:271 weakly excites
  10. L16:2177 weakly suppresses
  11. L23:1438 weakly suppresses
  12. L16:536 weakly excites
  13. L24:2859 weakly suppresses
  14. L11:2342 weakly suppresses
  15. L8:1253 weakly excites
  16. L3:1341 weakly suppresses
  17. L19:1566 weakly excites
  18. L13:2504 weakly excites
  19. L5:43 weakly excites
  20. L4:356 weakly excites
  21. L23:651 weakly suppresses
  22. L8:2048 weakly suppresses
  23. L10:2216 weakly excites
  24. L5:1704 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.

2 values between -3.07 and -2.971 values between -2.86 and -2.761 values between -2.76 and -2.661 values between -2.66 and -2.561 values between -2.56 and -2.461 values between -2.35 and -2.251 values between -2.05 and -1.951 values between -1.95 and -1.851 values between -1.74 and -1.641 values between -1.64 and -1.541 values between -1.44 and -1.342 values between -1.34 and -1.231 values between -1.23 and -1.131 values between -1.13 and -1.032 values between -1.03 and -0.933 values between -0.93 and -0.833 values between -0.83 and -0.734 values between -0.73 and -0.6212 values between -0.62 and -0.5214 values between -0.52 and -0.4222 values between -0.42 and -0.3255 values between -0.32 and -0.22214 values between -0.22 and -0.111265 values between -0.11 and -0.011894 values between -0.01 and 0.09310 values between 0.09 and 0.1966 values between 0.19 and 0.2918 values between 0.29 and 0.403 values between 0.40 and 0.502 values between 0.50 and 0.602 values between 0.60 and 0.702 values between 0.70 and 0.801 values between 0.90 and 1.01offnegativepositiveoften

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. sender
  2. lep
  3. ToSend
  4. hud
  5. lope
  6. préc
  7. Publisher
  8. pep
  9. *s
  10. .Make

Suppresses

  1. theast
  2. # ⏎
  3. bler
  4. 部份
  5. raises
  6. #⏎⏎
  7. ailability
  8. ใหญ
  9. aining
  10. みると