Neuron Explorer

Qwen3-0.6B

"the" after "percent of" or "a lot of"

A neuron in layer 3 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 "the" in phrases like "30 percent of the", "a lot of the" or "all the", where a quantity is followed by "the" and a noun. Its negative side shows up on parts of the word "alpine" and "Alpine", which is a different pattern. Turning it up did not produce a clear change in the model's 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

    the people Israel and responds. The plagues used to dislodge the Hebrew slaves are physical indicators of God's personal involvement. Admittedly, not all the Hebrew slaves were liberated from their miserable environment. Over the span of several hundred years

  2. #2 peak

    report suggests Hawaii will see nearly a one-meter rise in ocean levels by the end of this century. It predicts that more than 6,000 of the state’s buildings and 20,000 people across all of Hawaii

  3. #3 peak

    Mediterranean’s fish and mollusk species can be found. Nevertheless, these too are threatened by fishing, which sometimes uses illegal methods. Almost 30 percent of the world’s fish species are either extinct or facing extinction.⏎Visit the YAB Gallery

  4. #4 peak

    morphine activates it. When she gave morphine to mice that couldn't make beta-arrestin, they were still numb to pain, but a lot of the negative side effects of the drug were missing. They didn't build tolerance to the

  5. #5 peak

    developed a process to manage change⏎China’s Change: The Greatest Show on Earth is not a claim for Chinese exceptionalism, indeed quite the opposite. Almost all the 20 essential ideas it identifies are the same or very similar to

  6. #6 peak

    steps on the moon, meaning the United States had beaten Russia in the space race, but it also led to new changes back down on Earth.⏎“A lot of the things we developed to engage in the space race and space exploration eventually filtered down 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 first and the second. The first and the second. The first and the second. The first and the second. The first and the second. The first and the second. The first and the second. The first and the second.

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

    about how acid rain badly affected rivers with salmon (Salmo salar) and lakes with trout (Salmo trutta) and char (Salvelinus alpinus). Also from the Adirondack Mountains in northeastern USA came reports of

  2. #2 peak

    the cause of one of the botanical mysteries of the Burren as the Dense Flowered Orchid is native of the Mediterranean yet here growing alongside the (Alpine) Gentian and often with Mountain

  3. #3 peak

    clothed beneath with closely adpressed hairs, and bright yellow, somewhat tubular flowers, usually produced in fours.⏎C. DECUMBENS.—A charming alpine species, of low, spreading growth, bright-green three-parted leaves, and

  4. #4 peak

    organisms, but they are quite the opposite. My specific interests are related to species’ responses to environmental change, and alpine plants provide an incredible system to understand changes in range margins and community composition in response

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. L0:1075 weakly excites
  2. L2:115 weakly excites
  3. L2:111 weakly suppresses
  4. L2:1539 weakly suppresses
  5. L2:483 weakly suppresses
  6. L2:49 weakly suppresses
  7. L1:2401 weakly suppresses
  8. L2:1657 weakly excites
  9. L0:2790 weakly excites
  10. L1:1303 weakly suppresses
  11. L2:2692 weakly excites
  12. L0:191 weakly suppresses
  13. L0:1493 weakly excites
  14. L2:549 weakly suppresses
  15. L1:2423 weakly suppresses
  16. L0:1740 weakly excites
  17. L2:1036 weakly excites
  18. L0:212 weakly excites
  19. L1:861 weakly suppresses
  20. L2:1395 weakly excites
  21. L1:189 weakly suppresses
  22. L1:585 weakly excites
  23. L1:2268 weakly suppresses
  24. L2:2348 weakly excites

Feeds later layers

  1. L4:2754 excites
  2. L6:586 excites
  3. L7:224 excites
  4. L18:2621 weakly excites
  5. L4:731 weakly excites
  6. L10:2536 weakly excites
  7. L25:1485 weakly suppresses
  8. L14:854 weakly excites
  9. L23:2893 weakly excites
  10. L12:1364 weakly excites
  11. L9:284 weakly suppresses
  12. L13:2758 weakly suppresses
  13. L14:2871 weakly suppresses
  14. L17:2348 weakly excites
  15. L20:566 weakly excites
  16. L9:656 weakly excites
  17. L15:2944 weakly excites
  18. L19:1900 weakly excites
  19. L12:3015 weakly suppresses
  20. L11:233 weakly excites
  21. L8:1029 weakly excites
  22. L14:376 weakly excites
  23. L13:3024 weakly suppresses unclear: verbs and particles mid-phrase
  24. L19:832 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 -4.89 and -4.571 values between -2.33 and -2.011 values between -2.01 and -1.692 values between -1.37 and -1.051 values between -1.05 and -0.7378 values between -0.73 and -0.41670 values between -0.41 and -0.093066 values between -0.09 and 0.2367 values between 0.23 and 0.554 values between 0.55 and 0.873 values between 1.19 and 1.513 values between 1.51 and 1.831 values between 1.83 and 2.151 values between 2.15 and 2.471 values between 2.79 and 3.111 values between 4.07 and 4.391 values between 4.39 and 4.711 values between 5.35 and 5.672 values between 5.67 and 5.991 values between 6.31 and 6.631 values between 6.63 and 6.951 values between 7.59 and 7.91offnegativepositiveoften

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. odos
  2. enge
  3. OfWork
  4. iments
  5. lectron
  6. ],[
  7. ahun
  8. éné
  9. aceut
  10. )][

Suppresses

  1. formik
  2. Turks
  3. itori
  4. 岗
  5. …”⏎⏎
  6. .rank
  7. TKey
  8. 理会
  9. →renderer
  10. ictionary