Neuron Explorer

Qwen3-0.6B

uppercase words starting with O

A neuron in layer 3 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 gives weak responses to capital-letter words that begin with 'O', such as 'ORNL', 'OHIO' or 'OBLO'. The responses are small compared with the neuron's top cases elsewhere, and the pattern is not very clean. Turning it up did not change the writing in a clear way.

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

    ).⏎Regional and global biogeochemical dynamics data can now be located and acquired through a metadata search system at the ORNL DAAC.⏎The ORNL DAAC recently released six additional data sets from

  2. #2 peak

    tool can move linearly. In this way, the best cutting condition can be maintained on the entire path.<|endoftext|>OHIO DEPARTMENT OF AGRICULTURE⏎Division of Plant Industry – Ap

  3. #3 peak

    from the folks at Edutopia that shares what one teacher wishes she had known at the beginning of her career

  4. #4 peak

    in order to disassemble and fully reassemble the toy. It consists of an outer round layer (the container), a lid and 10 removable pieces. OBLO is colored with three primary colors and one secondary color to appeal to its audience

  5. #5 peak

    to keep the system’s data safe. And as ARPANET gave way to the purely academic world of the early internet, the idealistic graduate students who<|endoftext|>OF FATTY ACID METABOLISM⏎coma due to a disorder of

  6. #6 peak

    rophilic Substitution has several factors that affect its relative rates and this includes the following: substituents, solvent and temperature. Substituent like –OH and Br has an effect in electrophilic aromatic

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: the fact that, in the first place, the fact that, in the first place, the fact that... (the fact that) (1) the fact that... (the fact that) (1) the fact that... (1

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

    men heard the shells edging closer. “And then bang!—one of the big shells struck the ground right in front of us and we all went up in the air. But we all come down again.”⏎York was showered with honors

  2. #2 peak

    THC . Attempts to use them clinically have been unsuccessful, due to difficulty separating the desired properties from unwanted psychoactive effects. Illicit drug manufacturers seized on the mind-altering potential of these chemicals and began distributing synthetic cannabinoid compounds in the early

  3. #3 peak

    quirrels. Indeed, following arrival of Pine Martens, Red Squirrels are now recolonising some forests from which they had been absent<|endoftext|>THE RAATZ ROUNDUP⏎Submitted for use on April 22, 20

  4. #4 peak

    and launched the project "Je tuin intekenen en vergelijken" on the citizen science platform MijnTuinlab (My Garden lab).⏎The project

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. L2:7 suppresses
  2. L0:2969 excites
  3. L2:355 weakly excites
  4. L0:1301 weakly suppresses
  5. L2:49 weakly suppresses
  6. L0:425 weakly suppresses
  7. L1:55 weakly excites
  8. L0:1224 weakly suppresses
  9. L0:178 weakly suppresses
  10. L0:1184 weakly suppresses
  11. L2:87 weakly excites
  12. L1:13 weakly excites
  13. L0:1280 weakly excites
  14. L0:1072 weakly suppresses
  15. L2:2328 weakly suppresses
  16. L1:2119 weakly excites
  17. L1:1099 weakly excites
  18. L2:1607 weakly suppresses
  19. L0:3069 weakly suppresses
  20. L2:833 weakly excites
  21. L2:493 weakly suppresses
  22. L2:1559 weakly excites
  23. L2:1376 weakly excites
  24. L2:38 weakly excites

Feeds later layers

  1. L7:2095 excites
  2. L14:901 excites mid-word fragments in technical terms
  3. L5:1071 weakly excites
  4. L20:1037 weakly excites
  5. L5:2484 weakly suppresses
  6. L19:1320 weakly excites
  7. L20:374 weakly suppresses
  8. L14:1425 weakly excites
  9. L18:871 weakly excites
  10. L26:1420 weakly suppresses
  11. L4:2163 weakly suppresses
  12. L10:709 weakly excites
  13. L4:2535 weakly suppresses
  14. L26:1362 weakly excites
  15. L21:1312 weakly suppresses
  16. L18:1401 weakly suppresses
  17. L18:954 weakly suppresses
  18. L18:2565 weakly suppresses
  19. L5:1641 weakly suppresses
  20. L17:2031 weakly suppresses
  21. L11:854 weakly excites
  22. L23:2001 weakly excites
  23. L15:1130 weakly excites
  24. L20:1329 weakly suppresses

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 -1.89 and -1.801 values between -1.62 and -1.531 values between -1.26 and -1.171 values between -1.08 and -0.982 values between -0.80 and -0.713 values between -0.71 and -0.623 values between -0.53 and -0.448 values between -0.44 and -0.3527 values between -0.35 and -0.26160 values between -0.26 and -0.17534 values between -0.17 and -0.081315 values between -0.08 and 0.011202 values between 0.01 and 0.10475 values between 0.10 and 0.19147 values between 0.19 and 0.2818 values between 0.28 and 0.374 values between 0.37 and 0.462 values between 0.46 and 0.552 values between 0.64 and 0.731 values between 1.19 and 1.281 values between 1.64 and 1.73offnegativepositiveoften

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. iele
  2. öh
  3. alion
  4. rien
  5. 更何况
  6. omers
  7. nets
  8. życie
  9. olders
  10. ieux

Suppresses

  1. ==
  2. obvious
  3. thern
  4. ’aut
  5. .reporting
  6. 自
  7. ptive
  8. ADF
  9. gy
  10. 万千瓦