Neuron Explorer

Qwen3-0.6B

first digit of a year like 2020

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 responds to the first digit of a year, mostly the "2" in 2000s dates such as 2020 or 2017. It is weaker on the leading digit of other measurements. When turned up, the model's output changed topic and became more ordinary prose, while turning it down made the output a string of zeros.

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

    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 2

  2. #2 peak

    Solution : Proposed by UN⏎- India’s take on it⏎Reasons of Conflict:⏎- The renewed violence in the region started on May 6, 2021 when Palestinians protested against an anticipated decision of the Israeli Supreme Court over

  3. #3 peak

    3(3), pages 315-320.⏎- Angus Deaton & Olivier Dupriez, 2011.⏎"Purchasing Power Parity Exchange Rates for the Global Poor

  4. #4 peak

    the issue by modifying the ALA<|endoftext|>Note: I published a previous version of this article on The Change Blog (now called Possibility of Change) in 2011.⏎What is life? Is it the state of physical existence,

  5. #5 peak

    sur l’évolution des charges et produits de l’Assurance Maladie au titre de 2014 (loi du 13 août 20

  6. #6 peak

    requirements of the PCI Security Standards Council. Outdated browsers do not support the requirements of modern web applications and pose significant security risks. After May 1, 2017, this browser will no longer be supported and you may experience issues

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 0.50000000000000000000000000000000000000000000

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

    36 short tons)|⏎|Length:||205 ft (62.48 m)o/a|⏎|Beam:||33 ft (10.06 m)|⏎|Draught:||11.

  2. #2 peak

    New York factory, which was a one-piece brass measuring 33.5 meters (109’) tall. This flagpole is still the tallest in service at An

  3. #3 peak

    500 UTC (11:00 a.m), Humberto had weakened to tropical storm status, with maximum sustained winds of 65 mph (105 km/h).⏎On September 14, Humberto had reached post

  4. #4 peak

    gent plate margin between the African Plate and the Eurasian Plate. It is the tallest active volcano in Europe, currently standing 3,329 m (10,922 ft) high, though this varies with summit eruptions

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. L24:2703 strongly suppresses
  2. L23:2761 excites
  3. L24:2118 excites
  4. L24:2879 suppresses
  5. L22:1187 suppresses
  6. L22:1505 excites
  7. L23:1295 suppresses
  8. L24:1676 excites zeros in large round numbers
  9. L23:510 suppresses
  10. L23:929 excites
  11. L24:430 suppresses
  12. L24:1777 suppresses
  13. L23:2629 suppresses
  14. L22:909 excites
  15. L23:2924 suppresses
  16. L23:2222 excites
  17. L23:615 excites
  18. L23:950 suppresses
  19. L21:2233 excites
  20. L24:2757 excites
  21. L24:2618 suppresses
  22. L22:1095 suppresses
  23. L22:434 excites
  24. L21:242 suppresses 'very' before an adjective

Feeds later layers

  1. L27:1153 strongly suppresses
  2. L26:413 strongly suppresses
  3. L27:862 suppresses
  4. L27:153 excites the spot just before a historical number
  5. L26:848 suppresses
  6. L27:1202 suppresses
  7. L27:396 suppresses
  8. L26:682 excites
  9. L26:598 excites spot just before a year or century
  10. L26:1126 suppresses the digit 1 in powers of ten
  11. L27:1522 weakly excites
  12. L27:328 weakly excites
  13. L27:352 weakly excites
  14. L26:834 weakly excites
  15. L27:992 weakly excites
  16. L27:180 weakly suppresses
  17. L26:1824 weakly suppresses
  18. L26:1521 weakly suppresses
  19. L27:1111 weakly suppresses
  20. L26:724 weakly excites
  21. L27:3011 weakly excites
  22. L27:902 weakly excites
  23. L26:2255 weakly excites
  24. L27:189 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.

3 values between -13.65 and -11.643 values between -11.64 and -9.633 values between -9.63 and -7.623 values between -7.62 and -5.6212 values between -5.62 and -3.6120 values between -3.61 and -1.603365 values between -1.60 and 0.41480 values between 0.41 and 2.427 values between 2.42 and 4.433 values between 4.43 and 6.442 values between 6.44 and 8.442 values between 8.44 and 10.451 values between 12.46 and 14.471 values between 14.47 and 16.481 values between 20.49 and 22.501 values between 38.57 and 40.581 values between 64.68 and 66.69offnegativepositiveoften

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. hare
  2. piel
  3. ingers
  4. ethyl
  5. EntityType
  6. yro
  7. urance
  8. 界第一
  9. eterminate
  10. igar

Suppresses

  1. 0
  2. 0
  3. ₀
  4. lify
  5. tas
  6. 迈进
  7. 零
  8. zero
  9. newline
  10. 动力