Neuron Explorer

Qwen3-0.6B

capital D starting a split word

A neuron in layer 27 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 a capital D that starts a fragment of a longer word, such as the 'D' in 'DPFs', 'Domicile' or 'Drowning'. Turning it up or down made the model's writing wander or repeat, without a clear link to the letter.

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

    to remove soot from diesel emissions in 1978, just a few years after the famed catalytic converter substrate. But like many environmental technologies, DPFs were initially viewed as an unnecessary luxury since diesel engine manufacturers easily met the

  2. #2 peak

    choice continues until the acquisition of another domicile of choice.⏎1. Domicile of choice: acquire when abandon previous domicile, arrive in the new domicile

  3. #3 peak

    drug discovery for vaccines⏎- Addressing the challenges of applying micro/nano technology for COVID-19 diagnosis, vaccination, or therapy.⏎Xuexin Duan, School of Precision Instrument and Opto-electronics Engineering, Tianjin University

  4. #4 peak

    scream for help. They will be using every bit of energy they possess to try and get air, if they are even able to get their face above water. Drowning is often a silent killer. Be aware of an expression of panic or concern

  5. #5 peak

    see how Southern Ute is using natural resources to advantage the tribe. We think they're doing well as far as adapting with the times,” said Navajo member Dailan Long, whose family lives near the Navajo Mine that is close to

  6. #6 peak

    two spacecraft. This system is currently being incorporated into the design of Mini AERCam vehicle.⏎This work was done by Nathan Howard of Johnson Space Center and Hai D. Nguyen of Lockheed Martin Corp. For further information, access http://aerc

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: each individual causes even causes each causes even causes each causes even causes each causes even causes each causes even causes each causes even causes each causes even causes each causes even causes each causes even causes each causes even causes each causes even causes each causes even

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

    without knowing. That is easy. How you get to know is what I want to know.⏎The above quote also reminded me of the Voltaire quote: “Doubt is not a pleasant condition, but certainty is absurd.”<|endoftext|>What is

  2. #2 peak

    | Fallon, Kevin. "Gabby Douglas, Ryan Lochte: Why Families of America's Olympics Athletes Are Broke." The Daily Beast. Newsweek/Daily Beast, Web. 20 Aug. 2012⏎

  3. #3 peak

    due to the Aizoaceae (Ice Plant Family), Asteraceae (Daisy family) and Campanulaceae (Bellflower Family).⏎The activity

  4. #4 peak

    NBC), "Ally McBeal" (FOX), "That '70s Show" (FOX), "The Simpsons" (FOX), and "Dawson's Creek" (FOX). Taylor said these programs were chosen because they

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. L23:400 excites
  2. L10:180 excites
  3. L5:196 suppresses the 'D' in vitamin D
  4. L17:300 suppresses
  5. L0:233 excites
  6. L18:105 suppresses
  7. L21:1912 excites
  8. L22:1511 suppresses
  9. L19:526 suppresses
  10. L2:110 excites
  11. L20:200 excites
  12. L23:206 suppresses
  13. L1:773 excites
  14. L19:372 suppresses
  15. L4:1373 suppresses
  16. L26:581 excites
  17. L6:13 excites
  18. L20:1120 excites
  19. L26:1564 weakly excites
  20. L1:10 weakly excites
  21. L18:569 weakly excites
  22. L8:7 weakly excites
  23. L22:473 weakly suppresses
  24. L22:530 weakly suppresses

Feeds later layers

None: this is the last layer.

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 -20.84 and -20.261 values between -17.32 and -16.731 values between -11.43 and -10.851 values between -4.96 and -4.381 values between -4.38 and -3.791 values between -2.61 and -2.022 values between -2.02 and -1.4322 values between -1.43 and -0.85177 values between -0.85 and -0.263594 values between -0.26 and 0.33101 values between 0.33 and 0.923 values between 0.92 and 1.511 values between 1.51 and 2.102 values between 2.10 and 2.68offnegativepositiveoften

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. _literals
  2. fold
  3. tabpanel
  4. 啰
  5. 样的
  6. organizer
  7. específ
  8. 税务局
  9. specifically
  10. supporter

Suppresses

  1. oe
  2. onation
  3. ชอบ
  4. rew
  5. omination
  6. midfield
  7. yssey
  8. .controls
  9. anship
  10. iane