Neuron Explorer

Qwen3-0.6B

'I' starting an acronym or word fragment

A neuron in layer 22 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 a capital letter 'I' that begins a short abbreviation or a piece of a longer word, such as the 'I' in 'IUCN', 'ILO' or 'IBS'. Its effect on the written output was hard to see, because the steered text was unrelated to this pattern.

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

    edition of the International Basic Safety Standards. The edition is co-sponsored by seven other international organizations — European Commission (EC/Euratom), FAO, ILO, OECD/NEA, PAHO, UNEP and WHO.

  2. #2 peak

    security, and many other scientific domains.⏎"NVIDIA and IIT Delhi share the common vision of developing technologies that boost computing performance to exas

  3. #3 peak

    ), the Federal Institute for Drugs and Medical Devices (BfArM) and IFA. 03/31/2020.<|endoftext|>Def

  4. #4 peak

    . The study, conducted on Drosophila melanogaster by scientists at the Institute for Research in Biomedicine (IRB Barcelona) and led by ICREA researcher Marco Milán, has reproduced each of the steps known

  5. #5 peak

    the species.⏎The Great Slaty Woodpecker is a vulnerable species on the IUCN Red List, due to habitat loss. Without proper conservation measures, its

  6. #6 peak

    In studies, up to 75% of IBS sufferers have been helped with a low FODMAP diet.⏎FODMAP

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: I have to be able to create a right powerlomate. I were not able to create a powerlomate until the right powerlomate of the right side, and I were not able to create a powerlom

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

    political contention until it was taken under the control of the autocratic Roman emperors.⏎- List of Latin phrases⏎- Roman circus⏎- Idiot/idiocy<|endoftext|>Four European capital cities lie on Danube, Vienna, Bratisl

  2. #2 peak

    far stood outside the Standard Bank offices in Rosebank, Johannesburg, the Soweto Theatre in Soweto, the Iziko Museum in Cape Town, Maropeng in the Cradle of Humankind

  3. #3 peak

    ) It is derived from the Hebrew Ebion, "poor," and was not given, as Origen supposes, for their low view of Christ. Id. at 6

  4. #4 peak

    could be written on this subject. I'm guessing the "multi-million line production code base" in question is the Dropbox code base, and maybe Dropbox has an idiomatic way of writing Python with lots of "isinstance()"s.⏎> In

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. L18:205 excites
  2. L21:1492 excites
  3. L2:1463 suppresses
  4. L17:367 suppresses
  5. L20:2017 suppresses
  6. L1:45 excites
  7. L17:106 excites
  8. L10:1641 suppresses
  9. L19:551 excites
  10. L19:1241 excites
  11. L17:366 suppresses
  12. L20:342 suppresses
  13. L0:2262 excites
  14. L6:483 suppresses
  15. L16:503 suppresses
  16. L20:1228 suppresses
  17. L16:203 weakly excites
  18. L20:121 weakly excites
  19. L18:1442 weakly excites unclear: word fragments and digits inside numbers
  20. L0:47 weakly suppresses
  21. L1:985 weakly excites
  22. L1:1274 weakly excites
  23. L16:493 weakly excites
  24. L2:7 weakly excites

Feeds later layers

  1. L26:2485 excites capital I starting a word fragment
  2. L23:678 excites
  3. L24:2607 weakly excites
  4. L24:1116 weakly excites
  5. L27:2113 weakly excites
  6. L26:2469 weakly excites
  7. L25:785 weakly excites
  8. L26:1627 weakly suppresses
  9. L26:1959 weakly suppresses
  10. L26:1839 weakly excites
  11. L24:1468 weakly excites
  12. L23:2151 weakly excites
  13. L23:935 weakly excites
  14. L24:2219 weakly excites
  15. L25:2082 weakly excites first-person 'I' starting a clause
  16. L24:1972 weakly suppresses
  17. L23:2358 weakly excites
  18. L23:958 weakly suppresses
  19. L25:2583 weakly excites
  20. L27:732 weakly suppresses
  21. L24:420 weakly suppresses
  22. L24:2425 weakly excites
  23. L24:709 weakly excites
  24. L23:1494 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 -91.50 and -86.501 values between -81.50 and -76.501 values between -76.50 and -71.501 values between -36.50 and -31.502 values between -21.50 and -16.501 values between -16.50 and -11.502 values between -11.50 and -6.5033 values between -6.50 and -1.503858 values between -1.50 and 3.505 values between 3.50 and 8.501 values between 13.50 and 18.501 values between 23.50 and 28.501 values between 103.50 and 108.50offnegativepositiveoften

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. inis
  2. setting
  3. ampp
  4. Setting
  5. Setting
  6. believe
  7. Corinthians
  8. inion
  9. bero
  10. Dtype

Suppresses

  1. mes
  2. richt
  3. olate
  4. breakup
  5. olves
  6. net
  7. diff
  8. 侘
  9. nen
  10. 噔