Neuron Explorer

Qwen3-0.6B

'ur' and 'ud' after a capital H

A neuron in layer 2 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 the second piece of words that start with a capital H, such as the 'UR' in HURRICANE or the 'UT' in HUTCHINSON. Turning it up gave wandering text that did not mention weather or storms.

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

    them. (We’ll start saving). For now, his contribution to the national weather conversation is this Hurricane FAQ.⏎LET’S START SMALL. WHAT IS A HURRIC

  2. #2 peak

    If the cyclone keeps growing and sustains 74 mph winds, it graduates to hurricane.⏎ONCE WE CALL IT A HURRICANE, HOW DO WE CATEGORIZE IT?⏎We look to the

  3. #3 peak

    a typhoon if it happened in the northwestern Pacific.⏎WHAT’S THE DIFFERENCE BETWEEN A HURRICANE AND A TROPICAL STORM?⏎Wind speed. When tropical cycl

  4. #4 peak

    gasoline, if the bottles are stored in your garage) or from foods that have gone bad in your<|endoftext|>SHORT HISTORY OF STINNETT, TEXAS HUTCHINSON COUNTY⏎April 6th, 2010⏎

  5. #5 peak

    6 types of lepton, 3 of which are negative, the other 3 neutral.<|endoftext|>From: Nicholas Holton, River House, Inc., Huron

  6. #6 peak

    more than 400 varieties of biscuits and employed more than 5,000 people.⏎Huntley & Palmers employed the latest technology but their success was due more to inspired

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 problem of the problem is the same as the problem of the problem of the problem... It is the same as the problem. So, the problem of the problem of the problem. So. So. So. So. So.

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

    ons.(Figure⏎of Consensus Splice Sites) A majority of these searches are based on algorithms that look for complete genes.⏎The Gene Finder site from the⏎Baylor College of Medicine has a variety of options to search for introns and

  2. #2 peak

    Cooper.⏎Buy the hardcover edition of⏎I Heard the Willow Weep for $6.00. (Retail price

  3. #3 peak

    , it might be useful to assume that your search path for⏎cd starts from one of those folders.⏎For example, I spend a lot of time working in⏎/var/www. Instead of typing⏎cd /var/www/example.com all the time

  4. #4 peak

    appropriate office for investigation.<|endoftext|>Skip to main content⏎Create interactive lessons using any digital content including wikis with our free sister product⏎. Get it on the⏎Pages and Files⏎Democratic Teaching Methods⏎Hidden Histories, Dramatic Events⏎

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. L0:1301 weakly suppresses
  2. L1:2522 weakly excites
  3. L0:2788 weakly suppresses
  4. L0:1825 weakly suppresses
  5. L1:53 weakly excites
  6. L1:900 weakly excites
  7. L0:1224 weakly suppresses
  8. L1:2110 weakly suppresses
  9. L0:178 weakly suppresses
  10. L1:2683 weakly excites
  11. L1:1935 weakly excites
  12. L0:2879 weakly excites
  13. L0:1464 weakly excites
  14. L1:413 weakly suppresses
  15. L0:559 weakly suppresses
  16. L1:929 weakly excites
  17. L1:1335 weakly excites
  18. L1:2505 weakly suppresses
  19. L0:2095 weakly excites
  20. L0:134 weakly suppresses
  21. L0:1259 weakly excites
  22. L1:1304 weakly suppresses
  23. L0:2105 weakly excites
  24. L0:424 weakly suppresses open parenthesis after a name or term

Feeds later layers

  1. L5:1417 weakly suppresses
  2. L14:1969 weakly excites
  3. L6:2213 weakly suppresses
  4. L20:663 weakly excites
  5. L4:849 weakly excites
  6. L19:1968 weakly excites
  7. L10:2946 weakly excites
  8. L5:3039 weakly suppresses
  9. L25:1273 weakly suppresses
  10. L4:2931 weakly suppresses
  11. L19:2311 weakly excites
  12. L22:761 weakly suppresses
  13. L23:1495 weakly suppresses
  14. L10:1361 weakly excites
  15. L9:1379 weakly suppresses
  16. L13:1924 weakly excites
  17. L14:2807 weakly excites
  18. L24:1747 weakly suppresses
  19. L17:1019 weakly excites
  20. L8:3063 weakly excites
  21. L3:1318 weakly suppresses
  22. L16:457 weakly suppresses
  23. L24:23 weakly suppresses
  24. L20:1909 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 -1.62 and -1.541 values between -1.01 and -0.921 values between -0.83 and -0.743 values between -0.65 and -0.571 values between -0.57 and -0.482 values between -0.48 and -0.3913 values between -0.39 and -0.3091 values between -0.30 and -0.21383 values between -0.21 and -0.13901 values between -0.13 and -0.041556 values between -0.04 and 0.05674 values between 0.05 and 0.14223 values between 0.14 and 0.2333 values between 0.23 and 0.3211 values between 0.32 and 0.406 values between 0.40 and 0.491 values between 0.49 and 0.582 values between 0.58 and 0.671 values between 0.76 and 0.852 values between 0.93 and 1.021 values between 1.11 and 1.201 values between 1.82 and 1.90offnegativepositiveoften

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. })(
  2. pipes
  3. десят
  4. amaha
  5. riott
  6. сел
  7. DateTimeKind
  8. plers
  9. _resolver
  10. consect

Suppresses

  1. ように
  2. for
  3. far
  4. ratio
  5. far
  6. bağlantı
  7. LAY
  8. 鲤
  9. 脲
  10. edly