Neuron Explorer

Qwen3-0.6B

the middle digit of years like 2020

A neuron in layer 19 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 second digit of a year written as a number, especially years like 2020 where the model has to decide the next digits. Turning it up did not make the model write about years, so its effect on text is unclear.

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

    . You may even have heard of names like Bitcoin or maybe Ethereum to name a view. There are many most popular cryptocurrencies which can be used in 2020.⏎Maybe you’ve heard the terms digital gold or halving if you’ve

  2. #2 peak

    . So, you could also argue the Internet (<|endoftext|>このトピックには0件の返信が含まれ、1人の参加者がいます。2020年7月6日 at 9:45 AMに Megan Smith

  3. #3 peak

    legal rights to unionized or lack thereof).Please provide resources as required by APA standards.⏎https://mynursingpaper.com/wp-content/uploads/2020/06/cropped-155972031

  4. #4 peak

    This post contains my opinions and is not meant to be taken as medical advice. If you have a medical concern, please consult your personal physician.⏎@2020,<|endoftext|>The global energy system works on timescales of decades rather years.

  5. #5 peak

    171, China⏎*Authors to whom correspondence should be addressed.⏎Received: 2020-8-4 / Accepted: 2020-9-8 / Published: 2020-9-

  6. #6 peak

    for acyclic graphs of pointers.<|endoftext|>The Death And Life Of The Great Lakes Pdf⏎- and pdf⏎- Tuesday, June 1, 2021 2:11:00 AM⏎- 1 comment⏎

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 want to be able to make a very small amount of a solution that can be applied on the following lines of the screen by a single click on the screen. The following lines are on the screen, and the solution is located on the

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

    0:10:22 CEST 2010⏎On 9/9/10 4:39 PM, Baba wrote:⏎> In below code "

  2. #2 peak

    EST 2010⏎On Wednesday 24 November 2010 22:12:37, Russ Abbott wrote:⏎> Cool. I wasn't aware of that notation. It doesn't quite get to

  3. #3 peak

    did so far:⏎Originally Posted by MrFujin⏎Enter a 5-charcter string: hgout⏎String#1: tough

  4. #4 peak

    a kind of battery: conductive electrodes coated with a compound called polyanthraquinone, which has a natural chemical attraction<|endoftext|>From: Michael Barrie <mibarrieinterchange.ubc.ca>⏎Subject: Overt Mark

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. L7:28 excites
  2. L14:1074 weakly excites
  3. L10:439 weakly suppresses
  4. L3:255 weakly excites
  5. L14:598 weakly excites
  6. L4:1943 weakly suppresses
  7. L9:764 weakly suppresses
  8. L17:1646 weakly excites
  9. L9:382 weakly excites
  10. L5:376 weakly excites
  11. L13:2150 weakly excites
  12. L13:2769 weakly excites
  13. L2:2641 weakly excites
  14. L9:748 weakly excites
  15. L8:2938 weakly excites
  16. L18:1263 weakly excites
  17. L11:140 weakly excites
  18. L17:1505 weakly suppresses
  19. L8:1711 weakly suppresses
  20. L5:404 weakly suppresses
  21. L0:2824 weakly suppresses
  22. L7:2152 weakly suppresses
  23. L15:1714 weakly suppresses
  24. L4:15 weakly suppresses

Feeds later layers

  1. L23:48 weakly suppresses
  2. L23:2827 weakly suppresses
  3. L20:1431 weakly suppresses
  4. L21:78 weakly suppresses
  5. L21:2459 weakly excites
  6. L22:1435 weakly excites
  7. L27:3021 weakly excites capital 'L' starting a name
  8. L24:429 weakly suppresses
  9. L22:2895 weakly excites
  10. L23:1414 weakly excites
  11. L26:99 weakly suppresses
  12. L21:2094 weakly excites
  13. L23:2982 weakly excites
  14. L21:1886 weakly suppresses
  15. L22:2244 weakly suppresses
  16. L21:1410 weakly excites
  17. L23:542 weakly excites
  18. L27:2027 weakly excites
  19. L20:2025 weakly excites
  20. L23:1969 weakly excites
  21. L21:1503 weakly excites
  22. L20:684 weakly suppresses
  23. L22:1094 weakly excites
  24. L23:1925 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 -12.52 and -11.592 values between -6.92 and -5.982 values between -5.98 and -5.053 values between -5.05 and -4.124 values between -4.12 and -3.189 values between -3.18 and -2.2512 values between -2.25 and -1.3170 values between -1.31 and -0.383731 values between -0.38 and 0.5666 values between 0.56 and 1.492 values between 1.49 and 2.421 values between 2.42 and 3.361 values between 5.23 and 6.162 values between 8.03 and 8.961 values between 17.37 and 18.301 values between 23.91 and 24.84offnegativepositiveoften

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. чат
  2. 受
  3. /validation
  4. variant
  5. 服务平台
  6. ioned
  7. чив
  8. IVATE
  9. 佔
  10. /${

Suppresses

  1. wave
  2. oso
  3. ampo
  4. osa
  5. odu
  6. ugs
  7. 好吗
  8. kea
  9. anders
  10. [..