Neuron Explorer

Qwen3-0.6B

prepositions after passive verbs

A neuron in layer 15 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on small connecting words like 'by', 'through', 'from' and 'with' that come right after a verb or noun and introduce how something happens, such as 'enabled by', 'improved through' or 'transmitted by the bite of'. When turned up, the model's text got loose and wandering, without a clear shift.

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

    sometimes travel towards Earth where they are captured by the Earth’s magnetic field and guided towards the geomagnetic polar regions. On their way down these particles are slowed down by Earth’s atmosphere, which acts as a shield. These charged particles collide with gas

  2. #2 peak

    scientists are working with EMBL's spin<|endoftext|>Babesiosis is caused by parasites of the genus babesia, which are transmitted in nature by the bite of

  3. #3 peak

    Ways Remote Sensing Can Contribute to Conservation,” featured in the latest edition of the scientific journal Conservation Biology.⏎Addressing global questions requires global datasets that are enabled by satellite remote sensing; this paper highlights the way in which continuous observations of the

  4. #4 peak

    ichia coli (E. coli), and against Streptococcus canis, which can be passed from companion animals to humans.⏎What diseases can you get from dog saliva?⏎This review focused on the most important viral and bacterial zoonotic

  5. #5 peak

    .⏎Combinations And Pure Substances⏎<|endoftext|>Pricing in the sky⏎Many policy problems (from road congestion to water and electricity shortages) can be greatly improved through the proper application of a simple concept–pricing. In this USA Today article,

  6. #6 peak

    of stored honey during the winter months, which helps the bees produce body heat. Heat energy is produced by the oxidation of the honey, and circulated throughout the hive by the wing-fanning of worker bees . Note the diagram at right.⏎On warmer

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: what to do to avoid to avoid to not to to to to to to to to to to to to to to to to to to to to to to to to to to to to to to to to to to to to to to to

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

    Rebinding an Old Book Tutorial⏎Got an old book that has its pages coming apart and its stitches loosening up? Well, take out your supplies, and follow the rebinding process in this chapter.⏎Inspect your book; whether it is a

  2. #2 peak

    elerate the spread of the virus is being well informed about the coronavirus. This means that people should be well informed about the disease-causing virus and the ways through which it spreads the disease. The only easy and effective way to protect yourself from

  3. #3 peak

    Sikh Gurus) and 'Dwara (ਦੁਆਰਾ)' (gateway in Gurmukhi), together meaning 'the gateway through which the Guru could be reached'. Thereafter, all Sikh places of worship came

  4. #4 peak

    the subject isn't proof that they Achilles and Patrokles were anything other than heroic friends, models of what love looked like between warriors who faced death and depended on each other implicitly for survival. For me, the silence means they were exactly as

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. L12:3004 suppresses
  2. L12:1040 suppresses
  3. L14:2512 suppresses
  4. L12:2685 excites
  5. L12:1733 excites
  6. L14:1953 suppresses
  7. L14:709 excites
  8. L9:1465 excites
  9. L14:2180 excites
  10. L6:2880 suppresses
  11. L9:2390 excites
  12. L14:2972 excites
  13. L12:1088 excites
  14. L4:1219 suppresses
  15. L6:1857 excites "until" after a negated clause
  16. L10:1242 excites
  17. L6:1956 suppresses "for" introducing a purpose or source
  18. L6:1726 suppresses
  19. L14:2416 suppresses
  20. L11:1267 excites
  21. L14:2044 suppresses
  22. L13:1228 excites
  23. L13:2366 suppresses
  24. L9:1340 suppresses

Feeds later layers

  1. L16:1267 excites
  2. L18:2911 excites
  3. L16:702 suppresses
  4. L16:1458 suppresses
  5. L16:1717 suppresses
  6. L16:1516 suppresses
  7. L19:597 excites
  8. L16:2157 excites
  9. L18:1502 excites
  10. L18:1509 suppresses
  11. L18:2903 suppresses
  12. L16:1216 suppresses
  13. L18:658 excites
  14. L17:484 excites
  15. L16:1948 excites
  16. L16:789 suppresses
  17. L16:2673 suppresses
  18. L18:290 suppresses
  19. L16:616 excites
  20. L17:863 excites
  21. L17:1241 excites
  22. L17:2840 excites
  23. L17:1957 suppresses
  24. L16:1531 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 -3.07 and -2.911 values between -2.13 and -1.981 values between -1.98 and -1.822 values between -1.82 and -1.662 values between -1.66 and -1.513 values between -1.51 and -1.354 values between -1.35 and -1.208 values between -1.20 and -1.0410 values between -1.04 and -0.8819 values between -0.88 and -0.7332 values between -0.73 and -0.5750 values between -0.57 and -0.41172 values between -0.41 and -0.26511 values between -0.26 and -0.101714 values between -0.10 and 0.06692 values between 0.06 and 0.21314 values between 0.21 and 0.37151 values between 0.37 and 0.5388 values between 0.53 and 0.6849 values between 0.68 and 0.8425 values between 0.84 and 0.9913 values between 0.99 and 1.1518 values between 1.15 and 1.317 values between 1.31 and 1.464 values between 1.46 and 1.625 values between 1.62 and 1.782 values between 1.78 and 1.932 values between 1.93 and 2.095 values between 2.09 and 2.242 values between 2.40 and 2.561 values between 3.03 and 3.18offnegativepositiveoften

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. express
  2. yp
  3. yal
  4. Express
  5. maté
  6. 一种
  7. 秀
  8. esk
  9. adies
  10. ayed

Suppresses

  1. NSS
  2. IGO
  3. OLS
  4. 谗
  5. NSE
  6. NEL
  7. ursed
  8. NSS
  9. ursal
  10. ventions