Neuron Explorer

Qwen3-0.6B

the letter d inside words

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 lights up on the letter d when it starts a word piece, like the d in dim, dementia, ditches or dimes. When it was turned up, the model's replies got stuck repeating filler words such as 'about about about', which does not clearly relate to the d pattern.

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

    ME.⏎DIM is traced to Old Norse dimmr (dark) andIEbasedhem (smoky). Russian smoke is dim; in⏎Turkish it is duman.⏎Edenic covers the silence and senselessness of DUMB. דמי

  2. #2 peak

    of all the different types of dementia (including younger-onset dementias), from the most-recognized - Alzheimer's - to the less-f

  3. #3 peak

    high on inside fence, and well-preserved section of ditch c 3m below rampart. On the SW and SE sides the defences consist of twin ditches with a medial rampart. The ditches are slight and on the SW

  4. #4 peak

    Accusative Nominative Accusative ichmich dudich erihnsiees wiruns ihreuchsie/Sie⏎Pre

  5. #5 peak

    όφῳ in Einklang zu bringen, ist es unumgänglich nothwendig dass der Ausdruck Roma quadrata technisch in doppeltem Sinne gebraucht wurde; einmal zur Be

  6. #6 peak

    700 pounds. Mechanical fastening is difficult because of a tendency for splitting. Predrilling for nails and screws is advisable.⏎Drying. The wood dries slowly with some risk of checking, end splitting, and warping. Slow

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:ismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismismism

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

    sons. He loved them very much. They always used to help one another<|endoftext|>Ramses III, Ramses also spelled Ramesses or Rameses (died 1156 bce, Thebes, Egypt), king of

  2. #2 peak

    trigger.” 26 U.S.C. § 5845(b). The machine gun was first widely used during World War I, where it “demonstrated its murderously effective firepower over and over again.” William Rosenau

  3. #3 peak

    with a specialist, rural emergency room physicians are more likely to adjust diagnosis and course of treatment for injured children, research shows.⏎The use of technology to link far-distant practitioners

  4. #4 peak

    combination of employer and employee contributions and offer a range of options, including Health Maintenance Organizations (HMOs), Preferred Provider Organizations (PPOs), and High-Deductible Health Plans (HDHPs

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. L5:196 suppresses the 'D' in vitamin D
  3. L10:180 excites
  4. L17:300 suppresses
  5. L0:233 excites
  6. L19:372 suppresses
  7. L22:1511 suppresses
  8. L2:110 excites
  9. L20:200 excites
  10. L20:1659 excites
  11. L1:10 excites
  12. L21:1912 weakly excites
  13. L18:105 weakly suppresses
  14. L1:773 weakly excites
  15. L23:206 weakly suppresses
  16. L22:473 weakly suppresses
  17. L26:1564 weakly excites
  18. L18:704 weakly suppresses
  19. L18:554 weakly suppresses
  20. L4:1828 weakly suppresses
  21. L19:441 weakly excites
  22. L18:569 weakly excites
  23. L16:461 weakly suppresses
  24. L19:526 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 -14.06 and -13.541 values between -13.03 and -12.511 values between -9.40 and -8.881 values between -6.81 and -6.291 values between -5.77 and -5.261 values between -4.74 and -4.221 values between -3.70 and -3.182 values between -2.67 and -2.153 values between -2.15 and -1.633 values between -1.63 and -1.1127 values between -1.11 and -0.59328 values between -0.59 and -0.083448 values between -0.08 and 0.4470 values between 0.44 and 0.968 values between 0.96 and 1.484 values between 1.48 and 2.001 values between 2.00 and 2.521 values between 2.52 and 3.031 values between 3.03 and 3.551 values between 3.55 and 4.071 values between 4.07 and 4.591 values between 5.11 and 5.622 values between 6.14 and 6.66offnegativepositiveoften

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. دد
  2. daf
  3. dad
  4. gable
  5. ться
  6. 和发展
  7. Collection
  8. mega
  9. 也应该
  10. guna

Suppresses

  1. ism
  2. ifer
  3. iss
  4. ove
  5. edic
  6. isting
  7. immers
  8. icism
  9. edb
  10. 记者了解