Neuron Explorer

Qwen3-0.6B

're' prefix meaning again

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

Lights up on the 're' piece at the start of a verb that means doing something again, such as recharge, reprogram, reconfigure or rebase. Turning it up made the model repeat words and phrases, which is not tied to the trigger.

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

    process isn’t perfect though, so these batteries lose total charge capacity after hundreds of recharge cycles. Fortunately, some newer options can sustain a strong charge through thousands of recharges.⏎

  2. #2 peak

    versatile and can quickly be relocated to another part of your production process. ndeed, compared to larger complex solutions such as traditional robots, cobots are quick to reprogram, lightweight to move around and can easily be relocated as necessary. This is

  3. #3 peak

    the /dev directory. The devfsadm command also maintains the path_to_inst instance database.⏎Updates to the /dev and /devices directories in response to dynamic reconfiguration events or file system accesses are handled by devfsadmd, the daemon

  4. #4 peak

    into a logistical nightmare, requiring great amounts of time and creating high operations costs. Terraform is an effective solution to this issue, allowing administrators to rapidly deploy and reconfigure infrastructure all from one place, and all at the same time. It is

  5. #5 peak

    Nabokov learned over many years to do for his readers, fits with neuroscience’s recent understanding of brain plasticity, the degree to which the brain can be retrained, fine-tuned, redeployed.<|endoftext|>According to a study by

  6. #6 peak

    iffs provide useful information about your repository in a legible format, making it easier to perform the following actions:⏎- Review changes while staging.⏎- Decide whether to rebase or merge.⏎- Copy and paste code from a file.⏎Giving you even

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: building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building building

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

    He attended church services even in the face of stiff opposition. He refused to be initiated into the Ogboni cult (a traditional secret society) as a pre-requisite to his installation.⏎According to the available historical sources it was an admirable feat

  2. #2 peak

    ” scientists in much the same way that cold fusion (Low Energy Nuclear Reactions) have also been widely ridiculed by hot fusion academics.⏎Yet Shawyer

  3. #3 peak

    - Alternate day fasting⏎- 5/2 fasting, which involves eating regularly for 5 days, with 2 non-consecutive fast days.⏎- Time-restricted fasting, which limits eating within a certain window of time each day. This

  4. #4 peak

    uterium and tritium (isotopes of hydrogen) into plasmoids - plasma contained by a magnetic field through a process called FRC (field-reversed configuration). It magnetically accelerates the plasmoids down long tapered tubes

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. L21:2349 excites
  2. L20:3033 suppresses
  3. L20:112 suppresses
  4. L1:951 weakly excites
  5. L6:18 weakly excites
  6. L19:233 weakly suppresses
  7. L6:726 weakly suppresses
  8. L23:206 weakly suppresses
  9. L26:495 weakly suppresses
  10. L22:2646 weakly excites
  11. L16:306 weakly excites
  12. L18:569 weakly excites
  13. L0:2398 weakly suppresses
  14. L22:2630 weakly suppresses
  15. L8:16 weakly suppresses
  16. L11:2 weakly suppresses unclear: blank before a year or number
  17. L1:571 weakly excites
  18. L9:1548 weakly excites
  19. L7:102 weakly suppresses
  20. L3:236 weakly excites
  21. L0:3045 weakly excites
  22. L23:1461 weakly excites
  23. L11:429 weakly suppresses
  24. L1:853 weakly excites

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 -31.63 and -30.601 values between -26.48 and -25.452 values between -22.36 and -21.331 values between -19.27 and -18.241 values between -16.18 and -15.151 values between -13.09 and -12.061 values between -10.00 and -8.972 values between -7.94 and -6.912 values between -5.88 and -4.852 values between -4.85 and -3.822 values between -3.82 and -2.792 values between -2.79 and -1.7617 values between -1.76 and -0.733650 values between -0.73 and 0.30207 values between 0.30 and 1.3311 values between 1.33 and 2.361 values between 2.36 and 3.383 values between 6.47 and 7.501 values between 8.53 and 9.56offnegativepositiveoften

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. dogs
  2. ramid
  3. ths
  4. ständ
  5. 产出
  6. riminator
  7. avia
  8. MethodName
  9. 罔
  10. -output

Suppresses

  1. _unref
  2. anymore
  3. 反而
  4. reff
  5. pill
  6. artment
  7. lat
  8. scarce
  9. lại
  10. building