Neuron Explorer

Qwen3-0.6B

the word 'help' in calls to action

A neuron in layer 24 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 lights up on the word "help" when it appears in phrases like "You can help" or "Need help", especially in website notices and boilerplate text. When it was turned down, the model kept coming back to the word "help" over and over.

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

    regular classes.⏎If you experience problems downloading a file, check if you have the proper application to view it first. In case of further problems read the IDEAS help page. Note that these files are not on the IDEAS site. Please be

  2. #2 peak

    sum of money to support them and they both are euqally friendless in the town of Brussels. Will chance once again intervene? Will life<|endoftext|>You can help keep your heart and blood vessels healthy by taking steps toward a healthier lifestyle. These

  3. #3 peak

    )⏎- Sendust properties⏎|This alloy-related article is a stub. You can help Wikipedia by expanding it.|<|endoftext|>What national legislation governs cultural property issues?⏎The

  4. #4 peak

    carbon dioxide takes place.⏎Last Editorial Review: 6/14/2012⏎Back to MedTerms online medical dictionary A-Z List⏎Need help identifying pills and medications?⏎Get the latest health and medical information delivered direct to your

  5. #5 peak

    (Series 3) Zoology 1 (6): 199–204.⏎|This shark-related article is a stub. You can help Wikipedia by expanding it.|<|endoftext|>Sorry, fashion industry: The

  6. #6 peak

    : Planum temporale|⏎|Approximate location of Wernicke's area highlighted in gray|⏎| This article needs additional citations for verification.|⏎Please help improve this article by adding reliable references. Unsourced material may be challenged and removed

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 have a very large online market (or more specifically, a very large online market for private IP (or IP domain names or other IP-based domain names) or more specifically, a very large IP domain or IP domain name or IP IP

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 Truth⏎Grades 3 – 12 | Student Interactive | Organizing & Summarizing⏎The Persuasion Map is an interactive graphic organizer that enables students to map out their arguments for a persuasive essay or debate.⏎-

  2. #2 peak

    into your daily life, and managing some of the practical things, such as getting the right shoes or working with a personal trainer.⏎This guide’s main goal is to help you become more active, get the most from your activities, and do them

  3. #3 peak

    shape, and affects both females and males as well as children, youth and adults.⏎Our eating disorders program provides a range of services in a safe and supportive environment to both youth and adults, and their families, who are dealing with anorexia

  4. #4 peak

    is to teach students that it is O.K. to ''act autistic'' and also how to get by in a world where it is not.⏎Trained in self-advocacy, students proudly recite the positive traits autism can

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. L22:2167 weakly suppresses
  2. L0:893 weakly excites
  3. L16:1057 weakly suppresses
  4. L21:1140 weakly excites
  5. L21:954 weakly suppresses
  6. L22:919 weakly excites
  7. L4:398 weakly excites
  8. L22:1166 weakly suppresses 'to' after words needing an infinitive
  9. L11:428 weakly excites
  10. L19:2535 weakly excites
  11. L14:1304 weakly excites
  12. L2:2335 weakly excites
  13. L18:1058 weakly excites
  14. L5:611 weakly excites
  15. L20:1473 weakly suppresses
  16. L21:1914 weakly excites
  17. L23:2824 weakly excites
  18. L2:1066 weakly suppresses
  19. L22:1822 weakly suppresses
  20. L19:2264 weakly suppresses
  21. L16:1619 weakly suppresses
  22. L9:1690 weakly suppresses
  23. L1:2241 weakly excites
  24. L19:1809 weakly excites

Feeds later layers

  1. L27:1478 strongly suppresses
  2. L25:1516 suppresses
  3. L27:2531 suppresses
  4. L25:2519 excites
  5. L25:1777 weakly excites
  6. L25:86 weakly excites
  7. L25:2423 weakly excites
  8. L25:285 weakly suppresses gap right before a number
  9. L27:391 weakly excites
  10. L26:2515 weakly suppresses
  11. L27:1736 weakly suppresses
  12. L27:3018 weakly suppresses
  13. L26:1833 weakly excites
  14. L25:1942 weakly excites
  15. L25:2493 weakly excites
  16. L25:1999 weakly excites
  17. L27:2462 weakly suppresses
  18. L25:2629 weakly excites 'scripting' before 'language', language loops
  19. L26:739 weakly excites
  20. L25:960 weakly suppresses
  21. L27:2501 weakly excites
  22. L27:1204 weakly suppresses
  23. L26:2687 weakly excites unclear: word starts like Ag and Re
  24. L26:1768 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.

2 values between -17.92 and -16.063 values between -10.48 and -8.622 values between -8.62 and -6.754 values between -6.75 and -4.897 values between -4.89 and -3.0355 values between -3.03 and -1.173673 values between -1.17 and 0.69125 values between 0.69 and 2.5525 values between 2.55 and 4.414 values between 4.41 and 6.283 values between 6.28 and 8.142 values between 13.72 and 15.581 values between 23.03 and 24.891 values between 28.61 and 30.471 values between 54.67 and 56.53offnegativepositiveoften

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. 工委
  2. Publisher
  3. 圜
  4. 包袱
  5. 委
  6. Publisher
  7. irector
  8. setContent
  9. lok
  10. center

Suppresses

  1. help
  2. 帮助
  3. Help
  4. helping
  5. help
  6. 幫助
  7. helps
  8. helped
  9. (help
  10. Help