Neuron Explorer

Qwen3-0.6B

'both' after a paired subject

A neuron in layer 11 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

The neuron lights up on the word 'both' when it comes after a pair of things being discussed, like 'anchovies and sardines both' or 'the cathode and anode are both'. Turning it up made the model's text drift into unrelated rambling, so the steering results don't show a clear effect.

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 omega-3 fatty acids than anchovies, but both are good sources of beneficial fatty acids. Due to their small size, anchovies and sardines both are lower in mercury than larger fish. However, anchovies are a source of

  2. #2 peak

    the negative connects to an anode. The cathode and anode are both known as electrodes. They occupy most of the battery, and it’s here where

  3. #3 peak

    patient, love is understanding, and love is forgiving.⏎Which is correct sentence?⏎In order for a sentence to be grammatically correct, the subject and verb must both be singular or plural. In other words, the subject and verb must agree with

  4. #4 peak

    at least to Physicists!) experiment, Galileo went to the top of the leaning tower of Pisa and dropped a wooden ball and a lead ball, both the same size, but different masses.⏎- They both hit the ground at the

  5. #5 peak

    species. One sample was submerged in 96 percent ethanol for seven days, and the other in boiling distilled water for one day—both traditional preparations of plant extracts. The solvents were evaporated to complete dryness

  6. #6 peak

    the model can produce, with OM36 prototypes only being able to play 36 chords, however the OM36 and OM84 production models can both play 84 chord types. The OM84 was the first model to

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: the best time to start with the first of the month, the first of the month, and the first of the month, and the first of the month, and the first of the month, and the first of the month, and 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

    -Darwinist theory claims, has ever been found anywhere in the world despite the most diligent and prolonged search in the fossil record.⏎In their

  2. #2 peak

    was the ultimate moment of truth in a 350-year-long crisis and everybody knew it. What she did demanded a response, and everyone knew that, too. In 1955, it was not about polls or partisan politics

  3. #3 peak

    get others to join in a project or activity?⏎At UMass Dartmouth, we have designed programs, both inside and outside of the classroom, that we think will help you to become an educated person

  4. #4 peak

    museum was able to build a portrait collection following an international public appeal between 1917 and 1920 which was advertised in both the press and ration books.⏎Diane Lees, director general at the IWM, said

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. L4:262 strongly excites
  2. L8:126 excites
  3. L3:2187 suppresses
  4. L4:3031 excites
  5. L10:281 suppresses
  6. L3:2034 excites
  7. L5:2953 excites
  8. L6:1877 suppresses
  9. L9:2242 suppresses
  10. L2:1539 suppresses
  11. L7:1556 weakly excites
  12. L5:2610 weakly excites
  13. L7:2879 weakly excites
  14. L2:975 weakly suppresses
  15. L5:1500 weakly suppresses 'of' after a number or quantity
  16. L10:2023 weakly suppresses
  17. L2:1120 weakly suppresses
  18. L5:980 weakly suppresses
  19. L6:2358 weakly suppresses
  20. L10:1018 weakly suppresses
  21. L2:1865 weakly suppresses
  22. L6:66 weakly suppresses
  23. L2:803 weakly suppresses
  24. L5:2190 weakly excites

Feeds later layers

  1. L14:2551 excites
  2. L18:1800 excites
  3. L14:2463 excites
  4. L14:2371 excites
  5. L15:1678 excites
  6. L16:2394 suppresses
  7. L18:2450 excites "all" after a comma in a trailing phrase
  8. L13:2823 excites
  9. L15:2726 excites
  10. L18:2478 excites
  11. L16:1955 excites
  12. L13:2797 excites
  13. L12:312 suppresses
  14. L12:2970 excites
  15. L13:1278 weakly excites
  16. L16:2212 weakly excites
  17. L13:2554 weakly excites
  18. L18:590 weakly excites
  19. L16:1009 weakly excites
  20. L22:1974 weakly excites
  21. L18:2595 weakly excites 'only' in 'not only ... but also'
  22. L15:1781 weakly excites
  23. L16:1420 weakly excites
  24. L14:2546 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 -5.38 and -4.991 values between -3.03 and -2.641 values between -2.25 and -1.861 values between -1.86 and -1.463 values between -1.46 and -1.0710 values between -1.07 and -0.68568 values between -0.68 and -0.292958 values between -0.29 and 0.10207 values between 0.10 and 0.4965 values between 0.49 and 0.8929 values between 0.89 and 1.2825 values between 1.28 and 1.6712 values between 1.67 and 2.063 values between 2.06 and 2.454 values between 2.45 and 2.846 values between 2.84 and 3.236 values between 3.23 and 3.633 values between 3.63 and 4.022 values between 4.02 and 4.411 values between 4.41 and 4.801 values between 5.98 and 6.371 values between 9.89 and 10.28offnegativepositiveoften

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. 成员单位
  2. 纳税
  3. foil
  4. 불구하고
  5. 下雨
  6. 要考虑
  7. plode
  8. わけではない
  9. (fun
  10. ходят

Suppresses

  1. 两类
  2. głów
  3. adows
  4. рак
  5. 两个
  6. chief
  7. ⎞
  8. iest
  9. ucker
  10. obel