Neuron Explorer

Qwen3-0.6B

'of' in 'sort of' hedges

A neuron in layer 0 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 responds to the word 'of' in phrases like 'a sort of' or 'some sort of', where it introduces a loose, hedged kind of thing. Its strongest negative responses come on 'of' after verbs or nouns like 'warns of' or 'mood of'. Steering pushed the model into repetitive text rather than hedged phrases.

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 tree convert the wood into a sort of humus as they eat their way through the tree, creating debris that is soft

  2. #2 peak

    and a few other tribes living in the plains of Assam unsuccessfully sought a separate Union Territory, a sort of state directly governed by federal government. In 1987, the B

  3. #3 peak

    old school student with a⏎strong interest in mathematics.⏎This article gives an introduction to mathematical induction, a powerful method of mathematical proof.⏎An introduction to the sort of algebra studied at university, focussing on groups.⏎An introduction to some beautiful

  4. #4 peak

    treat those caused by<|endoftext|>Anorexia may be a disorder more of the metabolism than the mind, according to a new paper that argues the disease is a sort of cousin of diabetes.⏎But this theory of anorexia as a fundamentally biological disorder

  5. #5 peak

    back to society. . . .⏎Communitarianism cannot be classified as being left or right, and many claim to represent a sort of radical middle. Liberals in the American sense or social democrats in the European sense

  6. #6 peak

    manner."⏎Every time I've seen some sort of backup / recovery plan fail, it's usually a valid fail - ure, meaning

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: of the of the of the of the of the of the of the of the of the of the of the of the of the of the of the of the of the of the of the of the of the of the of the of 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

    four decades of newspaper opprobrium commenced. Over the years it has been blamed for poor mental health and numerous suicides and accidental deaths.⏎Today the government advice warns of the possibility of terrifying trips, vivid flashbacks and triggers for those susceptible to mental

  2. #2 peak

    existence of the perverse or transgressive.⏎Paying particular attention to Stevenson’s descriptions of the city at night, discuss how Stevenson uses descriptive passages to evoke a mood of dread.⏎At various junctures in Dr. Jekyll and Mr. Hyde

  3. #3 peak

    -access copy available on the SPIE Digital Library.<|endoftext|>Scientists at the University of Aberdeen are working to develop a system that could see smartphones used to warn epilepsy sufferers of the likelihood of a seizure before it happens.⏎Applied mathematicians from the University’s

  4. #4 peak

    pretensions to a Syro-Palestinian hegemony.⏎Two more years of peace ensued, but in Ramses’ 11th year a new coalition of Libyan tribes infiltrated the western delta. Compelled to wage yet another war

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

None: this is the first layer.

Feeds later layers

  1. L1:341 excites
  2. L4:4 suppresses
  3. L6:357 suppresses
  4. L19:715 suppresses
  5. L17:317 suppresses 'of' after a noun phrase
  6. L16:1088 suppresses 'of' after a noun phrase, as in 'of course'
  7. L22:229 suppresses "of" after a noun in "X of Y"
  8. L17:385 suppresses
  9. L18:293 suppresses
  10. L26:148 weakly suppresses
  11. L16:2412 weakly suppresses
  12. L1:571 weakly excites
  13. L18:90 weakly suppresses
  14. L15:286 weakly suppresses
  15. L7:6 weakly suppresses
  16. L5:5 weakly suppresses digit after a label like paper or type
  17. L17:1507 weakly suppresses
  18. L18:93 weakly suppresses
  19. L1:105 weakly excites
  20. L1:1752 weakly excites
  21. L26:310 weakly suppresses
  22. L17:1835 weakly suppresses
  23. L27:2058 weakly excites
  24. L27:2164 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 -3.07 and -2.981 values between -2.88 and -2.792 values between -2.79 and -2.701 values between -2.70 and -2.601 values between -2.60 and -2.512 values between -2.51 and -2.411 values between -2.41 and -2.321 values between -2.32 and -2.223 values between -2.22 and -2.134 values between -2.04 and -1.943 values between -1.94 and -1.852 values between -1.85 and -1.753 values between -1.75 and -1.663 values between -1.66 and -1.576 values between -1.57 and -1.479 values between -1.47 and -1.386 values between -1.38 and -1.288 values between -1.28 and -1.197 values between -1.19 and -1.0914 values between -1.09 and -1.0011 values between -1.00 and -0.919 values between -0.91 and -0.818 values between -0.81 and -0.7217 values between -0.72 and -0.6225 values between -0.62 and -0.5327 values between -0.53 and -0.4417 values between -0.44 and -0.3434 values between -0.34 and -0.2550 values between -0.25 and -0.15276 values between -0.15 and -0.062494 values between -0.06 and 0.04706 values between 0.04 and 0.13112 values between 0.13 and 0.2329 values between 0.23 and 0.328 values between 0.32 and 0.413 values between 0.41 and 0.512 values between 0.51 and 0.602 values between 0.60 and 0.70offnegativepositiveoften

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. ctors
  2. 开了
  3. чив
  4. 很想
  5. 很漂亮
  6. 锯
  7. 蒸
  8. ialis
  9. 因为它
  10. 没必要

Suppresses

  1. äter
  2. cram
  3. ath
  4. ars
  5. satisfaction
  6. upper
  7. growth
  8. conscience
  9. FetchType
  10. input