Neuron Explorer

Qwen3-0.6B

"by" after a passive verb

A neuron in layer 6 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 fires on the word "by" when it names who experienced or received something, as in "the pain experienced by the patient" or "suffered by another column". It goes the other way on "that" or "does" after the word "impact". Turning it up did not give a clear theme in the writing.

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

    of how dentin hypersensitivity pain is transmitted suggests that pressure or ionic changes in the fluid flow that exists in the dentin tubules stimulates the pain experienced by the patient. This is often referred to as the “hydrodynamic theory.” Inside

  2. #2 peak

    static universe, light from other stars would not be shifted by the Doppler effect⏎- The gravitational force felt by all matter would be infinite, causing all matter to vaporize into nucleons⏎

  3. #3 peak

    the country’s Indigenous population. As the first article concluded, the harm experienced by Indigenous communities calls for an extensive process of reconciliation. However, previous damage is not

  4. #4 peak

    improves negative symptoms experienced by people with chronic schizophrenia.⏎In the UK, many doctors and psychiatrists are referring

  5. #5 peak

    the vultures started dying in huge numbers. Their populations fell by 97 per cent in one decade. - one of the steepest declines experienced by any bird species.⏎In 2003 the culprit was identified - an

  6. #6 peak

    the persecutions suffered by Jews and Judaism in exile was not the decimation of our numbers, but 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: the impact of the pandemic on the world, the pandemic, and the virus of the pandemic. It is also important that the pandemic, and the virus of the pandemic, has shaped the world. So, the effect of the pandemic on 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

    nutrition and anemia. Although alcohol in small doses is an appetite stimulant, larger amounts suppress hunger, which doubly deprives the body of nutrients.⏎What impact does alcohol have on food intake and eating habits?⏎Different studies have found varying linkages

  2. #2 peak

    only when it becomes exceptionally poor, such as during wildfires, but the benefits of improved air quality can be reaped year-round. To further explore the vast impact that air quality has on human and environmental health, particularly coupled with

  3. #3 peak

    ated to highlight the voices of women survivors of the clandestine centres, shedding light on the acts of gender-based violence they underwent during their detention, and the impact that it had in their lives after the recovery of democracy in 1983

  4. #4 peak

    Released, the ozone-depleted air mixes with air across the southern hemisphere, diluting ozone at mid-latitudes during summer.⏎The atmospheric concentration of ozone-depleting CFCs and halons has been falling since the adoption of the Montreal Protocol

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. L5:571 suppresses
  2. L4:1394 excites
  3. L5:2207 excites
  4. L4:2646 excites
  5. L4:2170 weakly excites
  6. L2:832 weakly excites
  7. L1:2740 weakly excites word fragments inside non-English words
  8. L5:1479 weakly excites
  9. L5:2745 weakly suppresses
  10. L4:2302 weakly excites
  11. L5:2187 weakly excites
  12. L3:1055 weakly suppresses
  13. L5:1091 weakly suppresses
  14. L4:197 weakly excites
  15. L0:3047 weakly excites
  16. L5:363 weakly excites
  17. L3:717 weakly suppresses
  18. L5:2890 weakly suppresses
  19. L2:1215 weakly excites
  20. L4:2587 weakly excites
  21. L5:2214 weakly suppresses
  22. L3:370 weakly excites
  23. L4:619 weakly suppresses
  24. L2:1591 weakly excites

Feeds later layers

  1. L12:2535 excites
  2. L7:304 excites
  3. L9:654 excites
  4. L16:2119 excites
  5. L17:2528 excites
  6. L9:1465 suppresses
  7. L8:2411 suppresses
  8. L18:1233 excites
  9. L18:270 excites
  10. L14:2840 excites
  11. L17:2525 excites
  12. L16:1099 excites
  13. L16:702 suppresses
  14. L8:685 suppresses
  15. L11:1569 excites
  16. L19:2840 excites
  17. L9:2587 excites
  18. L14:2508 excites
  19. L14:1953 suppresses
  20. L10:2412 excites
  21. L12:2685 excites
  22. L21:787 excites
  23. L15:1974 excites
  24. L11:1911 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 -5.11 and -4.851 values between -4.59 and -4.342 values between -2.53 and -2.271 values between -2.27 and -2.011 values between -2.01 and -1.753 values between -1.75 and -1.503 values between -1.50 and -1.245 values between -1.24 and -0.988 values between -0.98 and -0.7220 values between -0.72 and -0.46187 values between -0.46 and -0.202424 values between -0.20 and 0.061118 values between 0.06 and 0.31108 values between 0.31 and 0.5710 values between 0.57 and 0.833 values between 0.83 and 1.093 values between 1.09 and 1.354 values between 1.35 and 1.602 values between 1.60 and 1.861 values between 1.86 and 2.121 values between 2.38 and 2.641 values between 4.96 and 5.22offnegativepositiveoften

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. .TabStop
  2. listeners
  3. consumers
  4. icult
  5. NavController
  6. taxpayers
  7. acea
  8. users
  9. 观众
  10. youngsters

Suppresses

  1. ernal
  2. 酒精
  3. ';⏎⏎⏎⏎
  4. ');</
  5. eding
  6. '\
  7. ogen
  8. ;⏎⏎⏎⏎
  9. ';⏎⏎⏎
  10. apor