Neuron Explorer

Qwen3-0.6B

"of" after "many" or "which"

A neuron in layer 2 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 weakly on the word "of" when it follows "many" or "which", as in "many of the" or "which of the". Its firing is faint, and turning it up did not clearly change the 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

    ?⏎Lying almost entirely within the tropics, and equally to north and south of the equator creates favourable conditions for rich wildlife. Africa is home to many of the world’s most famous fauna in human culture such as lions‚ rhinos‚

  2. #2 peak

    in inhabited structures.⏎Our final image is of the Chesa Futura in St. Moritz, Switzerland, a futuristic ski chalet which was built using many of the same automated wood-shaping technologies as are being employed in the construction of Bright

  3. #3 peak

    iral drugs.⏎The study addresses four distinct research questions. Most volunteers will participate in multiple components of the study to answer these questions. The first component will examine which of two proven strategies is safer and more effective at preventing mother-to-child HIV transmission before

  4. #4 peak

    in the cytoplasm, to the ribosomes. There the information of the mRNAs is used for protein production (translation). The fundamental question was which of the two processes, i.e. transcription or translation, plays the dominant role in

  5. #5 peak

    spirit of the translation of the Nigerian word Liberia, which means Freedom.⏎Among the early settlers, along with many of the workers from the southern and northeastern states, was also a very large Baham

  6. #6 peak

    fresh approach of the writers of the Harlem Renaissance and the importance of the addition of African-American voices to literature. The Van Vechten Collection includes portraits of many of Hughes’s contemporaries in the movement,

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 a cafe in a room. I have a room with a ceiling that's a triangle. The ceiling has a triangle. The floor is a square. What is the area of the room? Let me think. I

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

    more aware of the prevention and cure of diseases. A high literacy rate also imparts necessary skills to the people which they can utilize to solve their problems. Unfortunately,

  2. #2 peak

    server. Examples stock feeds, system status, weather conditions or other monitoring applications. This follows an Observer (a.k.a. Publish/Subscribe) pattern where remote clients (Observers) register for updates from

  3. #3 peak

    . Prized for their simple beauty and the strong medicinal value of their hips, they include Rosa Rugosa (Japanese Rose), Sweetbriar (a.k.a. Shakespeare's Eglantine), and Rosa Blanda (Prairie

  4. #4 peak

    Nuclear Reactor as the Antineutrino Source<|endoftext|>Greek government-debt crisis⏎|Greek debt crisis|⏎Greek government debt crisis articles⏎The Greek government-debt crisis (also known as the Greek Depression in reference to the Great Depression) is

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. L0:117 strongly suppresses
  2. L1:259 excites
  3. L0:2090 weakly suppresses
  4. L0:141 weakly excites the word 'of' in 'middle of'
  5. L1:217 weakly suppresses
  6. L1:597 weakly excites
  7. L0:2355 weakly suppresses
  8. L1:868 weakly suppresses
  9. L1:705 weakly excites
  10. L0:1064 weakly suppresses
  11. L1:850 weakly suppresses
  12. L1:663 weakly excites
  13. L0:425 weakly excites
  14. L1:108 weakly excites
  15. L1:2170 weakly suppresses
  16. L1:951 weakly excites
  17. L0:740 weakly excites
  18. L0:490 weakly excites
  19. L1:1232 weakly excites ".com" in web addresses
  20. L0:2010 weakly suppresses
  21. L1:2288 weakly excites
  22. L0:1660 weakly excites
  23. L1:1599 weakly suppresses
  24. L0:1672 weakly suppresses

Feeds later layers

  1. L21:571 strongly excites
  2. L26:2490 strongly excites
  3. L26:1579 strongly excites
  4. L21:2500 strongly excites
  5. L20:777 strongly excites
  6. L23:1351 strongly suppresses
  7. L26:1374 strongly excites
  8. L26:130 strongly suppresses ellipsis cut-off in quotes
  9. L21:1808 strongly excites
  10. L23:2344 strongly excites
  11. L26:1179 strongly excites
  12. L21:625 strongly excites
  13. L5:621 strongly excites
  14. L21:2021 strongly suppresses the 4 in years 184x, 164x
  15. L19:1870 strongly suppresses
  16. L26:1776 excites
  17. L24:938 excites
  18. L27:1602 excites
  19. L25:956 excites
  20. L25:2877 suppresses
  21. L25:396 excites
  22. L23:438 excites
  23. L26:2849 suppresses
  24. L27:692 suppresses 'Ch' starting a proper name

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 -0.80 and -0.753 values between -0.65 and -0.607 values between -0.60 and -0.557 values between -0.55 and -0.4911 values between -0.49 and -0.4422 values between -0.44 and -0.3942 values between -0.39 and -0.3478 values between -0.34 and -0.29116 values between -0.29 and -0.24185 values between -0.24 and -0.19256 values between -0.19 and -0.13347 values between -0.13 and -0.08497 values between -0.08 and -0.031122 values between -0.03 and 0.02497 values between 0.02 and 0.07244 values between 0.07 and 0.12135 values between 0.12 and 0.1895 values between 0.18 and 0.2350 values between 0.23 and 0.2855 values between 0.28 and 0.3331 values between 0.33 and 0.3824 values between 0.38 and 0.4324 values between 0.43 and 0.4820 values between 0.48 and 0.543 values between 0.54 and 0.595 values between 0.59 and 0.648 values between 0.64 and 0.698 values between 0.69 and 0.745 values between 0.74 and 0.793 values between 0.79 and 0.842 values between 0.84 and 0.901 values between 0.95 and 1.001 values between 1.05 and 1.102 values between 1.15 and 1.201 values between 1.20 and 1.26offnegativepositiveoften

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. ive
  2. gether
  3. ly
  4. ся
  5. ers
  6. ful
  7. обходим
  8. 公安局
  9. IAL
  10. ial

Suppresses

  1. er
  2. e
  3. eration
  4. erin
  5. ería
  6. erne
  7. t
  8. erate
  9. ej
  10. erator