Neuron Explorer

Qwen3-0.6B

"each" in "each other" phrases

A neuron in layer 10 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 "each" when it is part of a phrase like "each other" or "against each other", where the next word is expected to be "other". Turning it up made the model's writing drift into repetitive or oddly phrased text, which does not clearly relate to this word.

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

    too thin, it will not dissipate heat correctly and be more prone to warping.⏎The brake rotor consists of two braking contact surfaces that must be parallel with each other, or else the brakes will pulsate when they are applied. In order

  2. #2 peak

    drop will equal the sum of the voltage drops measured across each resistor separately. This tells you that both the resistors' voltage drop waveforms are in-phase with each other, since they add simply and directly.⏎Measure voltage dropped across both capacitors

  3. #3 peak

    ix more than a few pages of paper to your fridge.⏎To see this for yourself, stick two fridge magnets together, back to back. They will slide against each other in one direction. If you slide them at right angles to that, they

  4. #4 peak

    .⏎Find the area of the shaded region created by the two overlapping⏎triangles in terms of a and b?⏎Position the lines so that they are perpendicular to each other.⏎What can you say about the equations of perpendicular lines?⏎ABCD is

  5. #5 peak

    adjacent coloured cells (vertically or horizontally, not diagonally) are connected, and a shape is a group of coloured cells that are directly or indirectly connected to each other.⏎The diameter of a shape is the greatest distance between any two of its

  6. #6 peak

    too fast for most telescopes to track anyway.⏎On the day or two immediately before docking and after undocking, the Shuttle and Station will appear to be chasing each other across the night sky.⏎After moonset, between 1 and 2

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: how to do the matter of factly speaking about the world, and then again for the matter of the world's sake, what- more than the thing, if not to say, say, what, if at all, what- more

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

    basic board with future potential built in - you can buy Arduino shields to do a wide variety of things - it doesn't have an Operating System (OS) per se. Arduino has a great deal of support for accessing the board and driving serv

  2. #2 peak

    them.⏎The exhibit tells the story of migrant farmworkers, their work, and the people who serve them, in a number of ways. Photographs, for the most part by the superbly skilled professional photographers who have been acknowledged previously, document

  3. #3 peak

    historical-linguistical issues connected with the research about the etymology of the name Kudara. Some solutions will be proposed, although not definitive for the time being, but susceptible of further reconsiderations and developments should other material relating to the

  4. #4 peak

    line. Children in these circumstances are without reliable access to food, let alone to the right sort of food to ensure healthy development of their ability to learn

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. L6:1463 strongly excites
  2. L6:1122 strongly suppresses
  3. L9:929 strongly excites 'each' in 'each other'
  4. L6:1299 strongly suppresses
  5. L7:663 strongly suppresses
  6. L4:117 suppresses
  7. L7:919 excites
  8. L6:1695 excites
  9. L3:687 excites
  10. L6:2016 excites
  11. L8:534 suppresses
  12. L8:2074 excites
  13. L5:1387 excites
  14. L1:706 suppresses
  15. L4:2227 excites
  16. L8:611 suppresses
  17. L9:846 excites
  18. L9:1945 suppresses
  19. L4:557 suppresses
  20. L4:1891 suppresses
  21. L6:809 suppresses
  22. L6:108 suppresses
  23. L6:837 suppresses
  24. L7:1746 excites

Feeds later layers

  1. L16:1059 strongly suppresses
  2. L16:592 suppresses
  3. L18:828 suppresses
  4. L18:564 suppresses
  5. L11:1384 suppresses
  6. L15:1662 suppresses
  7. L15:2665 suppresses
  8. L17:2822 suppresses
  9. L14:713 suppresses
  10. L19:1204 suppresses
  11. L20:3066 suppresses
  12. L18:708 suppresses
  13. L17:2196 suppresses
  14. L18:347 suppresses
  15. L14:1304 suppresses
  16. L21:360 suppresses 'such' introducing examples
  17. L17:1171 suppresses
  18. L15:1616 excites word fragments mid-word
  19. L18:769 suppresses
  20. L19:327 suppresses
  21. L13:2019 suppresses
  22. L17:124 suppresses
  23. L11:513 suppresses words inside idioms and fixed phrases
  24. L16:1750 suppresses

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 -5.62 and -5.261 values between -4.55 and -4.201 values between -4.20 and -3.841 values between -3.84 and -3.482 values between -3.48 and -3.131 values between -3.13 and -2.775 values between -2.77 and -2.426 values between -2.42 and -2.069 values between -2.06 and -1.7123 values between -1.71 and -1.3534 values between -1.35 and -0.9995 values between -0.99 and -0.64229 values between -0.64 and -0.281412 values between -0.28 and 0.071907 values between 0.07 and 0.43145 values between 0.43 and 0.7912 values between 0.79 and 1.1410 values between 1.14 and 1.501 values between 1.50 and 1.852 values between 1.85 and 2.213 values between 2.21 and 2.561 values between 2.56 and 2.922 values between 2.92 and 3.271 values between 3.63 and 3.991 values between 3.99 and 4.341 values between 5.05 and 5.411 values between 5.41 and 5.761 values between 8.25 and 8.61offnegativepositiveoften

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. inez
  2. stra
  3. 打印
  4. hyper
  5. duk
  6. piè
  7. stadt
  8. 提供更多
  9. ine
  10. pattern

Suppresses

  1. ately
  2. ishly
  3. ingly
  4. imately
  5. icularly
  6. atively
  7. 然是
  8. ược
  9. atically
  10. asion