Neuron Explorer

Qwen3-0.6B

'and' or 'the' joining two parallel items

A neuron in layer 18 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron is active on the small joining words, such as 'and' or 'or', that link two matching things in a pair or list, like 'scientist and mom' or 'humans, animals and the environment'. Turning it up made the model produce repetitive, circular sentences, which does not clearly show what the neuron does.

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

    of contrasting brilliant color and shade. Some galleries insist that artists stand alongside their works during shows, juxtaposing the art and the artist.⏎Quiz: What is juxtaposition?⏎- The act of placing two things

  2. #2 peak

    an epidemic in much of the world. This year’s theme, “Unite to End TB”, highlights the need for a concerted effort by the TB research community and groups of patients and their families to effectively solve the problems that allow TB to still

  3. #3 peak

    of society to address today’s increasingly complex environmental problems. And there has been frustration on both sides: researchers want to see the result of their efforts being used, and decision makers need relevant information to assist in planning and practice.⏎To help close that

  4. #4 peak

    Raimi⏎In this podcast, RFF’s Daniel Raimi gives a balanced and accessible view of oil and gas development, drawing on both scholarly literature and his t...<|endoftext|>|Full name||Noam Chomsky

  5. #5 peak

    first woman allowed to observe at Caltech’s Palomar Observatory. Rubin also was a mother to four children–proving that she could be both a scientist and mom.⏎Strange to think of a time when women literally weren’t allowed to study

  6. #6 peak

    of individuals with intellectual disabilities may be rooted in the terms used to describe the condition, others have maintained that stigma may be inherent in the condition itself, not in the label used. This study took advantage of a recent change in terminology introduced in Israel

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: what is the function of the function in the function's function, and what is the function's function's function? What is the function's function's function's function? The function is the function's function's function's function's function.

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

    the industrial cotton make use of heavy pesticides, insecticides, herbicides and other dangerous chemicals that are harmful to humans, animals and the environment. Organic fabrics, unlike industrial cotton does not release any toxic fumes

  2. #2 peak

    to live out the American Dream. I came with a dream that one day I would be judged not by my birth, social status, or ideological affiliation but by my abilities and virtues.⏎What I most wanted from my new country was an equal chance

  3. #3 peak

    For the past few months, most people have isolated themselves indoors, and the impacts are striking. The lack of connections with each other, with our communities, and with the outdoors is negatively impacting our health, our wellness, our sense of calm—and

  4. #4 peak

    Biquard, translated by Geoffrey Strachan (1966). Frédéric Joliot-Curie: the man and his theories. Eriksson. p. 129.<|endoftext|>IBM is becoming

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. L17:183 strongly suppresses
  2. L17:739 strongly suppresses
  3. L17:522 strongly suppresses
  4. L17:102 strongly excites
  5. L17:339 strongly excites
  6. L17:493 strongly excites
  7. L17:212 strongly excites
  8. L17:182 strongly excites
  9. L17:373 strongly suppresses words after comparisons and contrasts
  10. L17:91 suppresses
  11. L17:194 excites
  12. L17:604 suppresses
  13. L17:396 suppresses
  14. L17:536 excites
  15. L17:242 suppresses
  16. L17:495 excites
  17. L17:306 excites
  18. L17:319 excites
  19. L17:608 suppresses
  20. L17:329 excites
  21. L17:653 suppresses
  22. L17:275 suppresses
  23. L17:422 suppresses
  24. L16:594 excites

Feeds later layers

  1. L20:227 suppresses
  2. L19:2289 weakly suppresses
  3. L26:2899 weakly suppresses
  4. L25:873 weakly excites
  5. L22:545 weakly suppresses
  6. L19:590 weakly suppresses
  7. L21:998 weakly suppresses
  8. L19:960 weakly suppresses
  9. L25:27 weakly suppresses
  10. L23:2866 weakly excites
  11. L22:2659 weakly suppresses
  12. L19:1641 weakly excites
  13. L23:1818 weakly excites
  14. L19:2827 weakly suppresses
  15. L19:2176 weakly suppresses
  16. L25:2237 weakly suppresses
  17. L20:2541 weakly excites
  18. L21:843 weakly suppresses
  19. L23:2639 weakly excites
  20. L26:888 weakly excites
  21. L26:1957 weakly suppresses
  22. L20:2535 weakly excites
  23. L20:1644 weakly suppresses
  24. L20:2986 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 -9.51 and -9.151 values between -4.90 and -4.551 values between -3.84 and -3.492 values between -3.49 and -3.133 values between -3.13 and -2.786 values between -2.78 and -2.4213 values between -2.42 and -2.0723 values between -2.07 and -1.7143 values between -1.71 and -1.3665 values between -1.36 and -1.01171 values between -1.01 and -0.65415 values between -0.65 and -0.302132 values between -0.30 and 0.06799 values between 0.06 and 0.41138 values between 0.41 and 0.7750 values between 0.77 and 1.1219 values between 1.12 and 1.4813 values between 1.48 and 1.834 values between 1.83 and 2.181 values between 2.18 and 2.544 values between 2.54 and 2.892 values between 2.89 and 3.252 values between 4.31 and 4.66offnegativepositiveoften

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. 人身
  2. TAIL
  3. 户籍
  4. ervations
  5. idental
  6. -labelledby
  7. astos
  8. roid
  9. rist
  10. 户口

Suppresses

  1. enary
  2. ikon
  3. nhạc
  4. wick
  5. _cache
  6. postpone
  7. 华尔
  8. utes
  9. лад
  10. shall