Neuron Explorer

Qwen3-0.6B

'of' after quantity or 'proliferation' nouns

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 fires on the word 'of' when it follows words like 'proliferation', 'hundreds', 'number' or 'table'. Turning it up did not produce a clear change in the writing's topic.

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

    highly specialised areas and distinct disciplines that may at times be interconnected, but that by and large ignore one another. There appears to be an overwhelming trend toward a proliferation of distinct

  2. #2 peak

    will tell if the pandemic’s pause afforded clinicians the opportunity to understand the benefits this technology can bring to their practice and their patients.⏎Recent decades have seen a proliferation of technological advances that promise to transform traditional methods of fabricating in-shoe foot orth

  3. #3 peak

    (when extensive land clearing began and exotic plants were first introduced from overseas). Hundreds if not thousands of plants fall into this category. Many native plants are very beautiful and are underused

  4. #4 peak

    human activity or carelessness. Agriculture suffers from highly elevated temperatures which diminish the yield in produce from useful cultivation. At the same time, these temperatures transport a proliferation of bad grasses and parasites. Any modification in the rain flow increases the probability of

  5. #5 peak

    Yes. It cannot be said too often that modern warfare threatens the very existence of human life on this planet.⏎Dr. Dyer argues that, with the proliferation of nuclear weapons, there is every reason to suppose that

  6. #6 peak

    for a beginner to understand than feeding tortoises in captivity. It is also true that no other subject is as riddled with misinformation and myths. The proliferation of inaccurate and often lethal advice on this subject in books, in magazine articles and on

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 am a student who has been to the school, and I am a student who has been to the school. But now, I am a student who has been to the school, and I am a student who has been to 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

    %. Women are slightly more likely to be e-cigarette users than men. The main reason given for using e-cigarettes was to aid in quitting smoking.Back to table of contents⏎Smoking is the leading cause of preventable death in Great Britain.

  2. #2 peak

    students can’t learn if they are distracted by hunger in the classroom, nor can they<|endoftext|>It's time for another tour of international food-borne and water-borne disease issues and outbreaks.⏎A snack food

  3. #3 peak

    ode45('rhs',tspan,y0)%call the routine ode45⏎plot(t,y) %plot the solution⏎The results give a table of values for t and y and a figure.⏎Next consider the following two ordinary differential

  4. #4 peak

    reproduced at the top of this page<|endoftext|>August 21, 2015⏎Station name board in old Minato Station unveiled.⏎To tell history of the Sanban Light Railroad which

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. L22:1871 weakly suppresses
  2. L25:2701 weakly excites
  3. L5:203 weakly excites
  4. L21:2621 weakly excites
  5. L7:601 weakly suppresses
  6. L27:2318 weakly excites
  7. L13:3048 weakly excites
  8. L25:1063 weakly suppresses
  9. L8:1231 weakly excites
  10. L24:593 weakly suppresses
  11. L27:542 weakly suppresses
  12. L17:553 weakly excites
  13. L10:1270 weakly excites
  14. L19:16 weakly suppresses
  15. L8:680 weakly excites
  16. L22:589 weakly suppresses
  17. L20:906 weakly suppresses
  18. L4:681 weakly excites
  19. L24:1864 weakly suppresses
  20. L9:21 weakly excites
  21. L8:729 weakly excites
  22. L9:1548 weakly excites
  23. L10:1419 weakly suppresses
  24. L7:741 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 -0.82 and -0.781 values between -0.55 and -0.512 values between -0.51 and -0.461 values between -0.46 and -0.422 values between -0.37 and -0.334 values between -0.33 and -0.286 values between -0.28 and -0.2413 values between -0.24 and -0.1936 values between -0.19 and -0.1589 values between -0.15 and -0.10240 values between -0.10 and -0.06790 values between -0.06 and -0.011909 values between -0.01 and 0.03500 values between 0.03 and 0.08159 values between 0.08 and 0.1277 values between 0.12 and 0.1735 values between 0.17 and 0.219 values between 0.21 and 0.267 values between 0.26 and 0.306 values between 0.30 and 0.353 values between 0.35 and 0.394 values between 0.39 and 0.443 values between 0.44 and 0.481 values between 0.48 and 0.532 values between 0.53 and 0.571 values between 0.57 and 0.622 values between 0.62 and 0.662 values between 0.66 and 0.712 values between 0.71 and 0.751 values between 0.93 and 0.98offnegativepositiveoften

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. ifest
  2. CreateUser
  3. 颈
  4. (Output
  5. ため
  6. iskey
  7. 遂
  8. ApplicationUser
  9. 乾隆
  10. (solution

Suppresses

  1. уществ
  2. cq
  3. рова
  4. abstract
  5. qa
  6. unate
  7. controlled
  8. agger
  9. раж
  10. aggio