Neuron Explorer

Qwen3-0.6B

'of' after a noun phrase head

A neuron in layer 24 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron reacts to the word 'of' that follows a noun in phrases like 'the factory of the future', 'technology of the time' or 'Rookie of the Year'. When turned up, the model's output became a long string of the word 'of', which fits what it pushes toward.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    and explore their potential impact on our lives.⏎Artificial Intelligence and Machine Learning⏎Artificial intelligence, or AI, has emerged as one of the most transformative technologies of our time. It encompasses machine learning, natural language processing, and computer vision,

  2. #2 peak

    was black. But the cost of producing much smaller batches of a wider variety, with each product tailored precisely to each customer’s whims, is falling. The factory of the future will focus on mass customisation—and may look more like those weavers

  3. #3 peak

    do to get past it.⏎The Pressure to Perform⏎Chuck Knoblauch was an outstanding baseball player in the 1990s. He won Rookie of the Year” and multiple Gold Gloves (outstanding defense). Because of his

  4. #4 peak

    December 21,1940) was an Irish American Jazz Age novelist and short story writer, who is generally regarded as one of the greatest American writers of the Twentieth Century. Fitzgerald's The Great Gatsby was perhaps the great

  5. #5 peak

    ⏎Front cover (photo)⏎Back cover (photo)⏎The Hughes Steamer⏎Like most young men, Howard Hughes was fascinated with automobiles. Unlike most other young men

  6. #6 peak

    New technologies often adopt the terms and design characteristics of the tech that came before them. So this wasn’t simply the television set of tomorrow, it was the radio set of tomorrow — now with TV!⏎This amazing bit of Depression-era futurism

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: one day in 2021, a single-fermerone field field field field field field field field field field field field field field field field field field field field field field field field field field field field field field field field field

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

    OF CARBOHYDRATE METABOLISM⏎due to an inborn error in the metabolism of carbohydrates occurs in fructose-1-6⏎diphosphate deficiency and in fructose-1-phosphate aldolase deficiency⏎(

  2. #2 peak

    After putting the drops in, keep the head tilted for several minutes to evenly distribute the medication within the ear canal.⏎Side Effects: Like all medications, Ciprofloxacin/Dexamethasone Ear Drops may cause side effects, although not

  3. #3 peak

    brew more sustainable beer. The company’s origins stem from the University of New South Wales where founders Constantine Tsounis, Khushal Polepalle and Bijil Subhash all met while studying.⏎“We’re chemical engineers but we’re also

  4. #4 peak

    bacteria, while dexamethasone is a steroid that helps reduce swelling and inflammation in the ear.⏎Dosage and Administration: The usual dosage of Ciprofloxacin/Dexamethasone Ear Drops is one or two drops into the

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. L21:1309 excites
  2. L11:1397 excites
  3. L20:224 suppresses
  4. L13:2971 weakly excites
  5. L9:1327 weakly excites
  6. L21:1442 weakly excites
  7. L8:1339 weakly suppresses
  8. L18:2103 weakly suppresses
  9. L18:569 weakly excites
  10. L16:981 weakly excites
  11. L5:703 weakly suppresses
  12. L18:2931 weakly suppresses
  13. L15:74 weakly excites
  14. L0:172 weakly excites
  15. L18:811 weakly excites
  16. L10:418 weakly suppresses
  17. L14:1279 weakly excites
  18. L5:646 weakly excites
  19. L20:3069 weakly suppresses
  20. L5:2047 weakly excites
  21. L11:2015 weakly excites
  22. L3:159 weakly excites
  23. L7:864 weakly suppresses
  24. L16:2689 weakly excites

Feeds later layers

  1. L26:711 strongly excites 'years' after an age number
  2. L27:447 strongly excites
  3. L25:2020 strongly excites
  4. L26:920 excites
  5. L26:279 suppresses
  6. L27:658 excites
  7. L26:115 suppresses
  8. L25:18 suppresses
  9. L27:1962 weakly excites
  10. L26:1320 weakly excites
  11. L25:2961 weakly suppresses
  12. L26:54 weakly excites
  13. L25:2263 weakly suppresses
  14. L27:556 weakly suppresses
  15. L26:750 weakly excites
  16. L25:2883 weakly excites
  17. L26:1519 weakly excites
  18. L27:1334 weakly suppresses
  19. L25:2840 weakly excites
  20. L27:815 weakly suppresses
  21. L26:1076 weakly excites
  22. L25:1862 weakly suppresses
  23. L25:2415 weakly suppresses
  24. L26:2154 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 -14.16 and -13.531 values between -11.02 and -10.391 values between -10.39 and -9.761 values between -9.76 and -9.132 values between -8.51 and -7.882 values between -7.88 and -7.252 values between -6.62 and -5.992 values between -5.99 and -5.374 values between -5.37 and -4.748 values between -4.74 and -4.1110 values between -4.11 and -3.483 values between -3.48 and -2.859 values between -2.85 and -2.2320 values between -2.23 and -1.6059 values between -1.60 and -0.97283 values between -0.97 and -0.342386 values between -0.34 and 0.29969 values between 0.29 and 0.9187 values between 0.91 and 1.5419 values between 1.54 and 2.1712 values between 2.17 and 2.805 values between 2.80 and 3.433 values between 3.43 and 4.054 values between 4.05 and 4.684 values between 4.68 and 5.312 values between 5.31 and 5.942 values between 5.94 and 6.572 values between 6.57 and 7.191 values between 7.82 and 8.451 values between 8.45 and 9.081 values between 9.71 and 10.332 values between 10.33 and 10.96offnegativepositiveoften

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. of
  2. _of
  3. Of
  4. Of
  5. of
  6. →of
  7. -of
  8. _OF
  9. _Of
  10. của

Suppresses

  1. ложения
  2. ��
  3. ifikasi
  4. 并对
  5. би
  6. ********************************************************************************
  7. laces
  8. 出来
  9. ilate
  10. ńsk