Neuron Explorer

Qwen3-0.6B

"of" in titles and set phrases

A neuron in layer 4 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 responds to the word "of" in fixed phrases such as "Table of Contents" and "Town of," and in phrases like "crisis of" or "level of." Turning it up didn't produce a clear change in topic in the sample outputs.

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

    3 | 240 pages | Cloth $55.00⏎American History / Women's Studies/Gender Studies⏎View main book page⏎Table of Contents⏎Chapter 1. Sleep Is for Sissies: Elite Males

  2. #2 peak

    suggested the name “Town of Stinnett” in honor of A.S. Stinnett, and this

  3. #3 peak

    Do not use your browser's "Refresh" button). Please email us if you're running the latest version of your browser and you still see this message.⏎Table of contents⏎Bluetooth is an industrial specification for wireless data transfer. Bluetooth connectivity is often

  4. #4 peak

    these jobs. Each chapter offers exercises, links to videos and websites, review quizzes, and suggested readings to learn more about the creation and production of theatre.⏎Table of Contents⏎- PRODUCTION AND BACKSTAGE ORGANIZATION⏎- PROD

  5. #5 peak

    ens back to a world-wide quasi-revolution in law from 100 years ago, called by many at the time, the solution to “a crisis of democracy.” Today, that approach may herald a new period of legal transition as recent

  6. #6 peak

    Symposium, his theory of intellectual development as the standard contemporary treatment of deductive reasoning is used as the context in which the contributors elaborate on their own perceptions.⏎Table of Contents⏎Contents: W.F. Overton, Competence and Procedures: Constraints

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 process is given as 1,2,3,4,5,6,7,8,9,10,12,13,14,15,16,17,18

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

    balancers.⏎The chemical name of pool "Stabilizer" is Cyanuric Acid and is commonly referred to by either name. The Stabilizer level of swimming pool water must be maintained at 30-50 ppm. The

  2. #2 peak

    , which is based on taste and odor thresholds. In fact, less than 1 percent of samples included in the USGS studies exceeded the EPA consumer advisory concentration of 20 micrograms per liter. Still uncertain, however, are possible human

  3. #3 peak

    of a fire, spill, or other hazard into the laboratory. Sliding sashes should not be removed from horizontal sliding sash hoods. The face opening of the hood should be kept small to improve the overall performance of the hood. If

  4. #4 peak

    . They are developing an electric car that will break – no, shatter – the existing E-1 class electric land speed record of 204 mph. By combining aerodynamics with new, cutting-edge green

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. L3:367 excites
  2. L2:568 weakly excites
  3. L1:2340 weakly suppresses
  4. L3:927 weakly suppresses
  5. L2:1599 weakly excites
  6. L2:3026 weakly suppresses
  7. L1:912 weakly suppresses
  8. L1:679 weakly suppresses
  9. L1:403 weakly excites
  10. L1:2993 weakly excites
  11. L2:803 weakly suppresses
  12. L1:2518 weakly excites
  13. L3:45 weakly excites
  14. L0:2744 weakly excites
  15. L2:995 weakly suppresses
  16. L1:919 weakly excites
  17. L3:477 weakly suppresses
  18. L2:501 weakly excites
  19. L2:1171 weakly suppresses
  20. L1:43 weakly suppresses
  21. L1:2353 weakly suppresses
  22. L0:1644 weakly excites
  23. L0:40 weakly suppresses 'of' in 'sort of' hedges
  24. L1:312 weakly excites

Feeds later layers

  1. L7:674 excites
  2. L6:136 suppresses
  3. L13:440 weakly excites
  4. L8:2782 weakly suppresses
  5. L23:39 weakly suppresses
  6. L8:419 weakly excites
  7. L11:399 weakly excites
  8. L18:3027 weakly suppresses
  9. L23:2784 weakly excites
  10. L9:130 weakly excites
  11. L17:1725 weakly suppresses
  12. L14:2198 weakly suppresses
  13. L16:1119 weakly excites
  14. L19:789 weakly suppresses
  15. L15:762 weakly excites
  16. L17:1160 weakly excites unclear: inside numbers and names
  17. L14:270 weakly excites
  18. L26:658 weakly suppresses
  19. L8:349 weakly excites 'of' inside title-like noun phrases
  20. L14:2863 weakly suppresses
  21. L15:2778 weakly excites
  22. L17:1686 weakly excites
  23. L5:550 weakly suppresses
  24. L6:1569 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 -1.70 and -1.621 values between -1.62 and -1.531 values between -1.53 and -1.441 values between -1.44 and -1.352 values between -1.35 and -1.266 values between -1.18 and -1.094 values between -1.09 and -1.004 values between -1.00 and -0.914 values between -0.91 and -0.827 values between -0.82 and -0.748 values between -0.74 and -0.656 values between -0.65 and -0.5621 values between -0.56 and -0.4717 values between -0.47 and -0.3844 values between -0.38 and -0.30113 values between -0.30 and -0.21278 values between -0.21 and -0.12684 values between -0.12 and -0.031597 values between -0.03 and 0.06701 values between 0.06 and 0.15252 values between 0.15 and 0.2372 values between 0.23 and 0.3232 values between 0.32 and 0.4115 values between 0.41 and 0.5011 values between 0.50 and 0.596 values between 0.59 and 0.674 values between 0.67 and 0.764 values between 0.76 and 0.854 values between 0.85 and 0.941 values between 0.94 and 1.031 values between 1.11 and 1.203 values between 1.29 and 1.381 values between 1.55 and 1.641 values between 1.73 and 1.82offnegativepositiveoften

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. special
  2. بار
  3. same
  4. arena
  5. "><!--
  6. ですよ
  7. uf
  8. c
  9. interval
  10. Wilmington

Suppresses

  1. YPD
  2. ادات
  3. orro
  4. 淖
  5. ponse
  6. enne
  7. ennes
  8. _Column
  9. .setColumns
  10. ủ