Neuron Explorer

Qwen3-0.6B

year right after 'in', 'by' or 'from'

A neuron in layer 25 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 fires on the spot just before a year is written, especially after words like "in", "by" or "from". It is anticipating the year, so it turns on when a date is about to come. Turning it up gave repeated number-like text, not clear sentences.

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

    the e-learning market in the United States is expected to grow $6.8 billion by 2015, up from $2.9 billion in 2010. According to Sanjeev Ahuja, the vice president

  2. #2 peak

    of Chemical Engineering⏎By most measures, MIT chemical engineering spinoff Verdox has been enjoying an exceptional year. The carbon capture and removal startup, launched in 2019, announced $80 million in funding in February from a

  3. #3 peak

    climate change. Even by California standards; it’s ambitious, calling for a 36% reduction in greenhouse gas emissions from 2005 levels by 2020 (statewide emissions decreased 5.3% between

  4. #4 peak

    have turned to designing their own chips for performance gains. Apple has used custom silicon for its iPhones and iPads since 2010—then, in 2020, it announced that it would design its own chips for Mac

  5. #5 peak

    in the metric tonne that are verifiable and tradeable. Bloomberg predicts that the price of carbon credits may rise by 3,000 percent by 2029 with better regulations.⏎The ever increasing demand for carbon credits could

  6. #6 peak

    ’ll be the first time the deficit has topped the $1 trillion mark since 2012. Deficits topped $1 trillion for four straight years from 2009 to 2012 in the aftermath of the Great

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 don't make any fraction of the fraction of fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction fraction

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

    lagellates of the genus Symbiodinium (zooxanthellae), similar to<|endoftext|>Defining Course

  2. #2 peak

    around the egg to make sure that it gets enough nourishment until it is mature enough to be fertilized by a sperm.<|endoftext|>Hives (known medically as urticaria) are smooth, raised, red, pink or white bumps of varying

  3. #3 peak

    1 diabetes.⏎© Children with Diabetes, Inc. 1995-2017. Comments and Feedback.<|endoftext|>Scoliosis vs. Ky

  4. #4 peak

    as nutritionists and dietitians all agree, a diet high in fiber including berries and other fruits are beneficial to maintain physical and mental health!⏎Blueberries and Raspberries counteract inflammation, which has been associated with cardiovascular disease and Alzheimer’s

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. L21:2497 strongly suppresses
  2. L20:391 suppresses
  3. L15:2013 excites
  4. L20:1089 excites
  5. L21:1815 suppresses
  6. L19:827 excites
  7. L17:1476 excites
  8. L19:2444 excites
  9. L16:2932 suppresses
  10. L19:234 suppresses
  11. L9:927 excites
  12. L24:969 excites
  13. L19:1583 suppresses
  14. L18:1108 suppresses
  15. L20:2472 excites
  16. L20:1666 suppresses
  17. L22:816 excites
  18. L20:1520 weakly excites
  19. L14:2494 weakly excites
  20. L21:1448 weakly excites
  21. L24:890 weakly excites
  22. L21:741 weakly suppresses
  23. L9:416 weakly suppresses
  24. L22:906 weakly suppresses

Feeds later layers

  1. L27:189 strongly suppresses
  2. L26:834 strongly excites
  3. L27:352 strongly excites
  4. L27:992 strongly excites
  5. L26:270 strongly suppresses
  6. L27:180 excites
  7. L26:598 suppresses spot just before a year or century
  8. L27:678 suppresses
  9. L26:848 suppresses
  10. L27:328 suppresses
  11. L27:132 suppresses
  12. L26:1083 excites
  13. L27:902 weakly suppresses
  14. L27:294 weakly suppresses
  15. L26:102 weakly excites
  16. L26:463 weakly suppresses digits inside years and numbers
  17. L27:205 weakly excites
  18. L27:428 weakly excites
  19. L27:530 weakly suppresses
  20. L27:364 weakly suppresses
  21. L26:580 weakly suppresses
  22. L27:1522 weakly suppresses
  23. L27:1440 weakly suppresses
  24. L27:435 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.

2 values between -7.75 and -5.381 values between -5.38 and -3.00158 values between -3.00 and -0.633629 values between -0.63 and 1.7573 values between 1.75 and 4.1222 values between 4.12 and 6.504 values between 6.50 and 8.878 values between 8.87 and 11.251 values between 13.62 and 16.002 values between 18.37 and 20.751 values between 23.12 and 25.501 values between 39.75 and 42.121 values between 42.12 and 44.501 values between 49.25 and 51.621 values between 61.12 and 63.501 values between 65.87 and 68.251 values between 70.62 and 73.001 values between 84.88 and 87.25offnegativepositiveoften

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
  2. 2
  3. ₂
  4. twenty
  5. zwe
  6. Twenty
  7. twenty
  8. -two
  9. two
  10. ²

Suppresses

  1. unan
  2. 틴
  3. oluşan
  4. warz
  5. ylon
  6. .stem
  7. asha
  8. ấy
  9. vrolet
  10. ução