Neuron Explorer

Qwen3-0.6B

the word 'systematic' in academic text

A neuron in layer 4 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 very strongly on the word 'systematic' (and 'methodically'), mostly in research and academic writing, such as 'systematic review' or 'systematic errors'. It hardly does anything in the other direction. Turning it up did not make the model's stories about this word, so the steering text is not informative here.

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

    conducted a systematic review of the research literature to determine which interventions improve cervical cancer prevention, screening,

  2. #2 peak

    by a systematic comparison of facts.⏎6. (n.) Hence: To carry on a process

  3. #3 peak

    The course is highly interactive and engaging. It brings a fresh perspective to classical material by focusing on developing two crucial logical skills: strategic construction of proofs and thesystematic search for counterexamples.⏎Concise multimedia lectures introduce each chapter of the course

  4. #4 peak

    1970s. Providing comparative analysis as well as reflection on the evolution of this important genre, Urban Chroniclers in Modern Latin America is the first systematic study of the Latin American writers who forged a new reading public in the early twentieth

  5. #5 peak

    amongst other damaged cells, they’re more likely to multiply and form a tumour.⏎The research team grew the organoids from mouse breast duct tissue. They methodically targeted single cells by switching on oncogenes known to cause cancer in humans.

  6. #6 peak

    offer a view inside the Earth. It was the geophysicists Wadati and Benioff, who in 1954 independently discovered the systematic arrangement of earthquakes in the places

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: this is a work of art. The idea of a plant. This is a poem about life. I think I could read about a plant. The idea of a tree. These are all about life. This is the heart. These are

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

    confession. Their charity to the poor may be said to exceed the proper bounds that prudence ought to set it, for it contributes to encourage great numbers of beggars, which are a great annoyance to the whole kingdom, and as I have often

  2. #2 peak

    said to have lost their lives here, mostly political prisoners, but also other "undesirables" such as Jehovah's Witnesses, Sinti or beggars.⏎

  3. #3 peak

    in schools and within the public. This is a analysis-oriented diploma that may be pursued by anyone excited by studying points in science education on the graduate stage, whether or not holding science teacher certification. The European Journal of Worldwide Legislation is firmly established as

  4. #4 peak

    explanation: "I don't think there has ever been evidence that if you improve blood glucose levels, you reduce mental illness symptoms, but if you go the other way and improve<|endoftext|>Mining companies already know their business is vulnerable to climate change legislation.

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. L3:2961 weakly excites
  2. L3:917 weakly excites
  3. L3:294 weakly excites
  4. L1:2749 weakly excites
  5. L2:1653 weakly excites
  6. L3:1830 weakly suppresses
  7. L3:1381 weakly suppresses
  8. L3:1371 weakly excites
  9. L3:899 weakly excites
  10. L3:2670 weakly suppresses
  11. L3:538 weakly excites
  12. L2:2224 weakly excites
  13. L1:1823 weakly suppresses
  14. L3:346 weakly excites
  15. L0:1624 weakly excites
  16. L2:1321 weakly suppresses
  17. L0:773 weakly suppresses
  18. L2:2871 weakly excites
  19. L2:461 weakly excites
  20. L2:521 weakly excites
  21. L3:2414 weakly excites
  22. L3:421 weakly suppresses
  23. L1:2907 weakly suppresses
  24. L2:561 weakly suppresses numbers right after 'these' or 'first'

Feeds later layers

  1. L8:725 excites
  2. L14:2680 weakly excites
  3. L9:2525 weakly excites 'and' joining paired adverbs
  4. L12:887 weakly suppresses
  5. L11:159 weakly suppresses
  6. L6:1917 weakly excites
  7. L20:2735 weakly suppresses
  8. L12:1859 weakly suppresses
  9. L23:311 weakly excites
  10. L10:1905 weakly excites
  11. L20:2208 weakly excites
  12. L11:1514 weakly excites
  13. L5:1327 weakly excites
  14. L5:615 weakly excites
  15. L8:2206 weakly suppresses
  16. L21:342 weakly excites
  17. L6:165 weakly excites
  18. L21:2455 weakly suppresses
  19. L16:2876 weakly suppresses
  20. L6:914 weakly excites
  21. L14:2762 weakly suppresses
  22. L7:1682 weakly suppresses
  23. L10:2058 weakly excites
  24. L9:1239 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 -1.13 and -0.952 values between -0.95 and -0.774 values between -0.77 and -0.5910 values between -0.59 and -0.4191 values between -0.41 and -0.23967 values between -0.23 and -0.062271 values between -0.06 and 0.12497 values between 0.12 and 0.3043 values between 0.30 and 0.488 values between 0.48 and 0.669 values between 0.66 and 0.841 values between 0.84 and 1.022 values between 1.02 and 1.191 values between 1.55 and 1.731 values between 5.83 and 6.01offnegativepositiveoften

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. 详细介绍
  2. 详细了解
  3. 宣传活动
  4. iou
  5. 质量管理
  6. memory
  7. cleanliness
  8. 彻底
  9. 全面落实
  10. 详解

Suppresses

  1. emple
  2. states
  3. assume
  4. よう
  5. abbrev
  6. hope
  7. sign
  8. arise
  9. aspire
  10. take