Neuron Explorer

Qwen3-0.6B

connectives inside parallel lists

A neuron in layer 15 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on small joining words like 'the', 'and' or 'to' that sit between two matching things, such as 'between Greenland and Iceland to the west and the United Kingdom'. It is weaker and less consistent on other text. When turned up, the model fell into loops of repeated phrases, so the steering does not show a clear theme.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    meetings. The geography that concerned us in the “Forgotten Waters” exercise described in this paper is the North Atlantic passage between Greenland and Iceland to the west and the United Kingdom to the east: the GIUK Gap, as it was known throughout

  2. #2 peak

    ichlids) > Pseudocrenilabrinae⏎Etymology: Sarotherodon: Greek, saros, -ou = sawdust + Greek, ther = animal * Greek, od

  3. #3 peak

    and work with the dignity befitting the importance of their task."⏎Each major program area is presented in a section, describing research, training, & product development; arts; health; and advocacy. Contents include a description of service programs for migrants

  4. #4 peak

    iii): The site of Al-Hijr bears unique testimony to the Nabataean civilisation, between the 2nd and 3rd centuries BC and the pre-Islamic period, and particularly in the 1st century AD. It

  5. #5 peak

    an evening with two renowned researchers—Richard Alley, Evan Pugh Professor of Geosciences and an Associate of the Earth and Environmental Systems Institute at Penn State, and Gavin Schmidt, Director of the NASA Goddard Institute for Space Studies—as they

  6. #6 peak

    1 northern hemisphere and 2011 southern hemisphere influenza seasons.⏎During weeks 29 to 30 (17 July 2011 to 30 July 2011), National Influenza Cent

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 have been reading about some people who are "not in the back of the seat to the end of the aisle to the front of the car, but in the middle of the back of the seat to the end of the aisle to 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

    intersect and perpendicular lines are the lines that intersect at 90 degrees.⏎Answer: Lines with the same slope are parallel and if the slope of one line is the negative reciprocal of the second line, then they are perpendicular.⏎Let's see how

  2. #2 peak

    mixture of a few longer items and some terms of 3 to 6 characters usually gives good results. Avoid words that overlap each other, with one being contained in the other, like

  3. #3 peak

    ⏎Case Western Reserve University⏎DNA in its native form is composed of two molecules that have the⏎characteristic of being complementary such that one strand associates⏎with the other in a

  4. #4 peak

    (ωt+ΦI) where Vmax, Imax are the peak values, ω is the common angular pulsation and ΦV, ΦI are the instant phase of each signal. The phase-difference can, therefore

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. L12:2808 suppresses
  2. L12:2346 excites
  3. L12:399 suppresses
  4. L12:925 suppresses
  5. L9:1449 suppresses
  6. L13:2888 excites
  7. L13:952 suppresses
  8. L13:622 excites
  9. L13:1577 excites
  10. L13:2525 excites
  11. L13:650 excites
  12. L14:706 excites
  13. L12:2662 excites
  14. L14:366 excites
  15. L14:2282 suppresses
  16. L12:116 suppresses
  17. L12:1505 suppresses
  18. L12:1937 weakly suppresses
  19. L13:1783 weakly suppresses
  20. L12:1212 weakly suppresses
  21. L14:2752 weakly suppresses
  22. L9:3012 weakly excites
  23. L6:1084 weakly suppresses
  24. L14:2054 weakly suppresses

Feeds later layers

  1. L17:1259 excites
  2. L16:2433 excites
  3. L16:1636 excites
  4. L17:1546 excites
  5. L16:2455 excites
  6. L16:2733 excites
  7. L17:2427 excites
  8. L18:2862 excites
  9. L16:492 suppresses
  10. L17:1880 suppresses
  11. L16:661 suppresses
  12. L16:767 suppresses
  13. L17:1133 excites
  14. L16:660 weakly excites
  15. L17:2002 weakly excites
  16. L17:1575 weakly excites
  17. L16:925 weakly excites
  18. L18:2313 weakly excites
  19. L16:553 weakly excites
  20. L19:392 weakly suppresses
  21. L19:2308 weakly excites
  22. L27:2556 weakly suppresses
  23. L16:2209 weakly suppresses
  24. L17:465 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 -1.82 and -1.711 values between -1.48 and -1.361 values between -1.13 and -1.024 values between -0.90 and -0.793 values between -0.79 and -0.676 values between -0.67 and -0.5514 values between -0.55 and -0.4457 values between -0.44 and -0.32221 values between -0.32 and -0.21541 values between -0.21 and -0.091383 values between -0.09 and 0.021044 values between 0.02 and 0.14422 values between 0.14 and 0.25117 values between 0.25 and 0.3728 values between 0.37 and 0.4817 values between 0.48 and 0.6014 values between 0.60 and 0.717 values between 0.71 and 0.836 values between 0.83 and 0.945 values between 0.94 and 1.065 values between 1.06 and 1.172 values between 1.17 and 1.291 values between 1.29 and 1.404 values between 1.40 and 1.521 values between 1.52 and 1.631 values between 1.63 and 1.751 values between 1.98 and 2.091 values between 2.67 and 2.78offnegativepositiveoften

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. 相反
  2. 来宾
  3. ').⏎
  4. npos
  5. ’aut
  6. ') ⏎
  7. NEY
  8. 'aut
  9. сообщает
  10. ) ⏎

Suppresses

  1. Grade
  2. Gr
  3. Gr
  4. within
  5. Grade
  6. grav
  7. grade
  8. ssc
  9. Gro
  10. .Gr