Neuron Explorer

Qwen3-0.6B

"of" in rules and codes

A neuron in layer 21 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 "of" in phrases like "rules of", "code of" or "set of rules", where a formal set of rules or a standard is about to be named. Turning it up made the model write about rules and codes of conduct in general terms.

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

    , which also contains information about 16 and 17 year olds, may be obtained from the Department of Enterprise and Employment. There is also a Code of Practice Concerning the Employment of Young Persons in Licensed Premises which a number of

  2. #2 peak

    another Crusade to Tunisia.⏎Credit: Catholic News Agency, https://www.catholicnewsagency.com/ Information herein posted under the "rules of fair use" to foster education and discussion in accordance with Section 107

  3. #3 peak

    uta-sutras, together with the Grihya-sutras (dealing with domestic ceremonies) and the Dharma-sutras (dealing with rules of conduct), make up

  4. #4 peak

    es Play Better Basketball⏎A Basketball term meaning to embarrass some one usually while slamming the ball over them.⏎James Naismith wrote the original 13 rules of basketball.<|endoftext|>a sheet with a printout of the biochemical analysis of blood can

  5. #5 peak

    hearing reflect to a certain extent Australian social values. Their purpose is to attempt to maintain an efficient court system and create justice before the law. Processes such as rules of evidence which ensure only the most applicable evidence is heard have been developed over many years

  6. #6 peak

    services provided by the Internet. The author notes that an AUP is the most important document a school will create since it states the terms, conditions, and rules of Internet use determined by a school or district. Topics include: What is an A

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 power of a single truth. The power of a single truth is often more powerful than the power of a truth power. Truth power is a concept that is often in the historical and political context of a country, often in the historical and

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

    Interpret and harness financial information in their decision making and strategic planning.⏎• Systematically assess the financial viability of their functions, plans and projects.⏎• Manage and control their budget effectively

  2. #2 peak

    all car models⏎Within the next five years, the EU is likely going to require OEMs to deliver full LCA analyses of all models, to demonstrate their environmental footprint. This will probably include a sustainability cap imposed on the total amount of carbon

  3. #3 peak

    system needs related to communicating with the public about well water quality.⏎Identify technology-based solutions that could help address gaps in the system.⏎Select most appropriate innovations based on financial, technical, organizational and political realities of stakeholder groups.⏎Disseminate

  4. #4 peak

    borrower's ability to repay a loan under challenging economic conditions, such as a rise in interest rates. In Hong Kong, stress tests are used to assess the borrower's ability to repay the mortgage

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. L16:2244 excites
  2. L12:1580 excites
  3. L8:648 weakly excites
  4. L15:1910 weakly excites
  5. L18:611 weakly excites
  6. L19:2222 weakly suppresses
  7. L20:238 weakly suppresses
  8. L14:2754 weakly excites
  9. L17:133 weakly suppresses
  10. L12:2505 weakly suppresses
  11. L1:2084 weakly excites
  12. L16:1479 weakly excites
  13. L20:3066 weakly suppresses
  14. L13:2530 weakly suppresses
  15. L19:2719 weakly excites
  16. L18:821 weakly excites
  17. L18:1883 weakly excites
  18. L7:932 weakly excites
  19. L3:860 weakly suppresses
  20. L4:882 weakly excites 'with' after provide or familiarize
  21. L14:1909 weakly excites
  22. L17:103 weakly excites
  23. L10:154 weakly excites
  24. L17:1363 weakly suppresses

Feeds later layers

  1. L26:1847 weakly excites
  2. L23:1143 weakly excites
  3. L23:1637 weakly suppresses
  4. L24:1921 weakly excites
  5. L27:2986 weakly excites
  6. L23:2678 weakly excites
  7. L23:2440 weakly excites
  8. L23:1542 weakly excites
  9. L22:864 weakly suppresses
  10. L22:387 weakly suppresses
  11. L23:1102 weakly suppresses
  12. L25:1516 weakly suppresses
  13. L24:594 weakly excites
  14. L24:2087 weakly excites
  15. L22:2220 weakly suppresses
  16. L25:279 weakly suppresses
  17. L26:1185 weakly suppresses
  18. L23:284 weakly excites
  19. L22:1011 weakly excites
  20. L22:739 weakly excites 'are' after 'here' introducing a list
  21. L23:2951 weakly suppresses
  22. L22:2922 weakly excites
  23. L22:2598 weakly suppresses
  24. L25:1703 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 -9.91 and -9.411 values between -6.94 and -6.451 values between -6.45 and -5.963 values between -5.46 and -4.972 values between -4.97 and -4.482 values between -4.48 and -3.981 values between -3.98 and -3.493 values between -3.49 and -2.9911 values between -2.99 and -2.5014 values between -2.50 and -2.0126 values between -2.01 and -1.5138 values between -1.51 and -1.02191 values between -1.02 and -0.531488 values between -0.53 and -0.031830 values between -0.03 and 0.46224 values between 0.46 and 0.9627 values between 0.96 and 1.4513 values between 1.45 and 1.947 values between 1.94 and 2.448 values between 2.44 and 2.934 values between 2.93 and 3.433 values between 3.43 and 3.922 values between 3.92 and 4.412 values between 4.41 and 4.911 values between 4.91 and 5.401 values between 5.40 and 5.891 values between 5.89 and 6.391 values between 6.39 and 6.881 values between 8.36 and 8.861 values between 9.35 and 9.84offnegativepositiveoften

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. 行为
  2. 的行为
  3. behavior
  4. behaviour
  5. behavior
  6. Behavior
  7. 行為
  8. comport
  9. behaviors
  10. Behaviour

Suppresses

  1. wis
  2. trag
  3. 毒素
  4. öst
  5. >true
  6. pragma
  7. 奥斯
  8. せい
  9. лив
  10. hower