Neuron Explorer

Qwen3-0.6B

'and' joining paired items

A neuron in layer 15 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 word 'and' when it links two matching things, like 'men and women', 'the cathode and the anode', or 'systolic and diastolic'. Turning it up made the model produce long strings of list-like repetition, not a clear theme.

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

    in the Western world at least, men and women do gain weight from the age of 35 onwards, averaging around half a kilo per year for men and a little more for women.⏎The reason is not an aging metabolism. While older

  2. #2 peak

    Blood pressure is one of the vital signs of life. It is normally expressed as two numbers, the first or the upper number indicating the systolic blood pressure, and

  3. #3 peak

    it is either shifted away by air movement or re-radiated by the glass surface.⏎Typically, a highly reflective material will have a low emissivity, and a dull darker colored material will have a high emissivity. Reducing the emiss

  4. #4 peak

    and regrettable risky behaviors. Once you turn 21, try sticking to the recommended daily limit of no more than two beers or glasses of wine for men and one for women.⏎Kick the bad habits!⏎Rub snuff, smoke cigarettes or

  5. #5 peak

    rechargeable batteries that you can use in place of disposables.⏎Instead, you might use nickel-cadmium batteries (NiCd). Nickel is the cathode and cadmium is the anode. The recharge process isn’t perfect though, so

  6. #6 peak

    antisense oligonucleotides (ASO) that modulate splicing for PK-M. The cells are visible under light microscopy in the left column, and the DNA in their nuclei shows up when using the blue dye DAPI in the

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: when I look at the problem of the "rational" and "rational" problem. Let's say we consider the problem of the "rational" and "rational" problem. Let's say that we consider the problem of

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

    ampere, but with different meanings. To avoid confusion, the prefix “ab-” was often added to cgs electromagnetic units (especially by Americans) and “stat-” to cgs electrostatic system units. Many workers did not use

  2. #2 peak

    was round and he set out to prove it. That's the kind of character that built America and the kind of character we can't afford to forget. I think it's great that we

  3. #3 peak

    and renamed HMS Racer. HMS Mentor was burned by the British at Pensacola in 1781 to prevent her capture by invading Spanish forces. HMS Racer foundered in a gale in the Gulf of Mexico in 1

  4. #4 peak

    . Kashmiris have played an integral role in Hindu civilization – Kalhana authored ‘Rajatarangini’, the first written history of Kashmir and contemporary Bharat; Sarangadeva helped evolve Carnatic music; scholars like Nagarjuna taught

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. L6:1121 excites 'while' joining two clauses after a comma
  2. L8:70 excites 'and' between stacked superlatives
  3. L12:925 suppresses
  4. L11:1595 suppresses
  5. L14:1187 suppresses
  6. L13:622 excites
  7. L6:912 suppresses
  8. L13:1021 weakly excites
  9. L12:399 weakly suppresses
  10. L9:2516 weakly excites
  11. L12:2485 weakly suppresses
  12. L13:1497 weakly suppresses
  13. L8:607 weakly excites
  14. L12:1937 weakly suppresses
  15. L14:440 weakly suppresses
  16. L14:2865 weakly suppresses
  17. L12:3067 weakly excites
  18. L12:1746 weakly suppresses
  19. L4:57 weakly suppresses
  20. L6:2945 weakly suppresses
  21. L9:92 weakly suppresses "and" or "or" joining two items
  22. L14:2851 weakly suppresses
  23. L13:933 weakly excites
  24. L13:503 weakly excites

Feeds later layers

  1. L18:2862 excites
  2. L16:1636 excites
  3. L17:2427 excites
  4. L17:1546 excites
  5. L16:2733 excites
  6. L17:2002 excites
  7. L17:1092 excites
  8. L17:1259 excites
  9. L17:1133 excites
  10. L18:619 excites
  11. L16:853 excites
  12. L18:2385 excites
  13. L16:2433 excites
  14. L17:595 excites
  15. L18:308 excites
  16. L16:661 weakly suppresses
  17. L16:1389 weakly excites
  18. L17:2701 weakly excites
  19. L17:1017 weakly excites
  20. L16:678 weakly suppresses
  21. L18:189 weakly suppresses
  22. L16:300 weakly excites commas inside lists of nouns
  23. L17:145 weakly suppresses
  24. L16:2200 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.

2 values between -2.23 and -2.052 values between -1.70 and -1.526 values between -1.52 and -1.345 values between -1.34 and -1.175 values between -1.17 and -0.9914 values between -0.99 and -0.8227 values between -0.82 and -0.6437 values between -0.64 and -0.4670 values between -0.46 and -0.29378 values between -0.29 and -0.111818 values between -0.11 and 0.071204 values between 0.07 and 0.24270 values between 0.24 and 0.4230 values between 0.42 and 0.6014 values between 0.60 and 0.779 values between 0.77 and 0.955 values between 0.95 and 1.122 values between 1.12 and 1.304 values between 1.30 and 1.481 values between 1.83 and 2.011 values between 2.01 and 2.181 values between 2.18 and 2.361 values between 3.77 and 3.941 values between 3.94 and 4.121 values between 4.65 and 4.82offnegativepositiveoften

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. Ariel
  2. 副主席
  3. credited
  4. eds
  5. -coded
  6. ectl
  7. ArrayType
  8. MSR
  9. 若要
  10. anked

Suppresses

  1. Tr
  2. Tr
  3. star
  4. ож
  5. 星
  6. ington
  7. star
  8. tr
  9. TX
  10. .setInt