Neuron Explorer

Qwen3-0.6B

'relevant' before a noun

A neuron in layer 8 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 "relevant" (and similar words like "matter" or "topic") when it comes right before the thing being talked about, like "the relevant qualifications". When it was turned up, the writing drifted into unrelated new topics, and turning it down made the text start rambling about a single idea.

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

    towards promoting the sport.⏎Swimming education is available in the USA. The department of Education is the largest employer of swimming instructors. Every instructor has to have the relevant qualifications to get employment in the program. The instructors also must have a valid National

  2. #2 peak

    Yet, Meier says, “I wasn’t too sure about the psychological effects. I hadn’t seen any research on the matter."⏎At the time, Meier was a graduate student, searching for a subject

  3. #3 peak

    transport make their fitness for transport questionable from a scientific point of view. In a review, the scientific knowledge about fitness for transport is reviewed against the background of the relevant European legislation.⏎Injury and unnecessary suffering⏎The European regulation (Council Regulation

  4. #4 peak

    activities at home and abroad, the Center for Combating Desertification in Arid and Semi-arid Areas (CCDASA) was launched recently to concentrate on relevant scientific researches on combating desertification supported by

  5. #5 peak

    developments in machine learning. Probability theory was employed to offer theoretical explanations as to why many machine learning models work (read Murphy’s book for an excellent book on the topic). Support vector machines offered an alternative to neural networks, based on learning

  6. #6 peak

    , and as a result, sends information directly to the brain via the production of neurotransmitters. The Shanghai University team therefore conducted a systematic review of 21 relevant studies to try and find out more about the potential effects of regulating the microbiota

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 more you think about it, the less you can see, the something in you. It's a process of thinking, which is the first, and the second, and third, and so on. But the something in you is a

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

    . Newer standards take a performance or program approach, identifying a safety or health objective and providing a framework for employers to use in achieving that goal. OSHA standards cover the gamut of workplace safety and health issues,

  2. #2 peak

    ige. Originally debuted at Carnegie Hall on January 23rd, 1943, the composition received mixed, if not critical feedback.⏎Much of that criticism weighed on the "pretentiousness" such a narrative composition implies. As

  3. #3 peak

    mechanical engineering, better enable its practitioners to contribute to the well-being of humankind.⏎ASME Core Values⏎In performing its mission, ASME adheres to these core values:⏎•Embrace integrity and ethical conduct⏎•Embrace diversity and

  4. #4 peak

    natural gas. Higher pressures result in coal and oil.⏎Leaking from natural gas pipelines contributes to some of the atmospheric green house gas contribution.⏎Other contribution arises from that organic breakdown that occurs from the natural fermentation process in landfills. Methane<|endoftext|>

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. L6:1135 suppresses
  2. L6:1218 suppresses
  3. L4:468 excites "that" after "than" or "like"
  4. L6:1458 suppresses
  5. L7:2537 weakly suppresses
  6. L5:2517 weakly suppresses
  7. L7:2482 weakly suppresses
  8. L4:2798 weakly excites
  9. L1:1456 weakly excites
  10. L5:2733 weakly suppresses
  11. L7:460 weakly suppresses
  12. L7:236 weakly suppresses
  13. L5:2376 weakly suppresses
  14. L2:974 weakly excites
  15. L3:2753 weakly suppresses
  16. L1:1084 weakly excites period in U.S. and initials
  17. L7:2650 weakly excites
  18. L4:270 weakly suppresses
  19. L5:610 weakly excites
  20. L0:2775 weakly suppresses
  21. L4:1733 weakly suppresses
  22. L6:1093 weakly suppresses
  23. L6:1999 weakly excites
  24. L2:2359 weakly suppresses

Feeds later layers

  1. L9:864 excites
  2. L9:175 excites
  3. L16:2660 excites
  4. L10:1103 excites
  5. L16:2861 excites
  6. L10:1389 excites
  7. L17:221 excites
  8. L11:1318 excites
  9. L12:1761 weakly excites
  10. L16:2466 weakly excites
  11. L15:2242 weakly excites
  12. L13:2601 weakly excites
  13. L9:82 weakly excites
  14. L12:1805 weakly suppresses
  15. L23:1119 weakly suppresses
  16. L11:2741 weakly excites
  17. L14:1956 weakly excites
  18. L10:376 weakly suppresses
  19. L17:1200 weakly suppresses
  20. L12:2962 weakly suppresses
  21. L9:1215 weakly excites
  22. L17:2011 weakly excites
  23. L21:1484 weakly suppresses comma after a sentence subject
  24. L14:2442 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 -1.68 and -1.572 values between -1.33 and -1.212 values between -1.21 and -1.091 values between -1.09 and -0.981 values between -0.98 and -0.862 values between -0.86 and -0.744 values between -0.74 and -0.626 values between -0.62 and -0.5122 values between -0.51 and -0.39118 values between -0.39 and -0.27469 values between -0.27 and -0.151110 values between -0.15 and -0.041526 values between -0.04 and 0.08462 values between 0.08 and 0.2097 values between 0.20 and 0.3228 values between 0.32 and 0.4413 values between 0.44 and 0.559 values between 0.55 and 0.6710 values between 0.67 and 0.794 values between 0.79 and 0.914 values between 0.91 and 1.023 values between 1.02 and 1.142 values between 1.14 and 1.262 values between 1.38 and 1.491 values between 1.49 and 1.612 values between 1.61 and 1.732 values between 1.85 and 1.972 values between 2.08 and 2.201 values between 2.44 and 2.551 values between 2.55 and 2.671 values between 2.91 and 3.02offnegativepositiveoften

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. íc
  2. 作出了
  3. .MoveNext
  4. 作出
  5. 其他玩家
  6. RenderWindow
  7. 的心理
  8. uation
  9. 作了
  10. uing

Suppresses

  1. uples
  2. resto
  3. ndef
  4. 맨
  5. System
  6. acie
  7. imitives
  8. тки
  9. _System
  10. ropy