Neuron Explorer

Qwen3-0.6B

'and' linking process words in lists

A neuron in layer 20 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 'and' in lists of actions or processes, such as 'gathering, managing, and analyzing data' or 'collection and analysis'. It is weakly negative on colons that introduce a list of steps or phases. The steering replies did not clearly reflect this.

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

    (GIS) is a framework for gathering, managing, and analyzing data. Incorporating GIS into the disaster response framework has improved planning, response

  2. #2 peak

    -polluting nations such as<|endoftext|>Learning disabilities are neurologically based conditions that interfere with the acquisition, storage, organization, and use of skills and knowledge. They are identified by deficits in academic functioning and in

  3. #3 peak

    to be successful, indigenous peoples must be seen as powerful agents of change, accorded access to decent work opportunities and the ability to participate in the development, implementation and evaluation of sustainable policies and measures aimed at combating climate change. Simultaneously,

  4. #4 peak

    , 1975. The Privacy Act is essentially an omnibus "code of fair information practices" that attempts to regulate the collection, maintenance, use and dissemination of personal information by federal executive branch agencies. These amendments to the FOIA

  5. #5 peak

    who don’t know what Analytics and Optimization mean. So here is a brief description for those who are new to this field.⏎Web Analytics is the measurement, collection and analysis, and reporting of internet data for the purpose of understanding and optimizing web usage

  6. #6 peak

    help detect, monitor, analyze and resolve the environmental problems, such as the contaminations of subsurface groundwater and soil

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: how to identify, assess, and design a plan for the identification of a proposal proposal proposal statement for a proposal about identification of a plan of planning for design, planning, planning identification, planning planning identification, planning planning identification, planning identification,

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

    to collaborate more efficiently and effectively. Emergency managers resolve natural disaster challenges using the Comprehensive Emergency Management (CEM) approach. CEM is broken down into four phases: mitigation, preparedness, response, and recovery (Agency, 199

  2. #2 peak

    push 3D printing forward, it’s helpful to remember where we’ve been. Stripped to essentials, the process of making an orthosis involves several steps:1⏎- An assessment of the patient’s medical condition and needs, including diagnosis

  3. #3 peak

    version of the C language (C11) and assembly language from scratch. It covers the entire path from source code to program execution, including generation of ELF object files, and static and

  4. #4 peak

    it will remain so. In fact, it’s an indispensable part of product development, especially in the software and manufacturing industries.⏎QA testing involves overseeing the production process from start to finish, with a focus on testing the product through its various phases to

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. L12:1651 suppresses
  2. L12:2590 suppresses
  3. L11:2165 suppresses
  4. L12:2912 weakly suppresses
  5. L6:2864 weakly excites
  6. L19:1126 weakly suppresses
  7. L19:2249 weakly excites
  8. L19:1288 weakly suppresses
  9. L12:55 weakly suppresses
  10. L16:2965 weakly excites
  11. L13:1845 weakly excites
  12. L13:1541 weakly suppresses
  13. L6:1684 weakly suppresses
  14. L8:2322 weakly suppresses
  15. L15:2634 weakly suppresses
  16. L8:1916 weakly excites
  17. L0:2485 weakly excites
  18. L19:285 weakly suppresses
  19. L19:2767 weakly suppresses
  20. L10:225 weakly excites
  21. L15:1942 weakly suppresses
  22. L8:1624 weakly excites
  23. L15:2221 weakly excites
  24. L17:1595 weakly excites

Feeds later layers

  1. L21:1651 suppresses
  2. L22:914 weakly suppresses
  3. L25:489 weakly excites
  4. L23:2458 weakly suppresses
  5. L22:1446 weakly suppresses
  6. L25:1294 weakly suppresses
  7. L22:1946 weakly suppresses
  8. L22:1185 weakly suppresses
  9. L24:967 weakly excites
  10. L23:836 weakly suppresses
  11. L21:1191 weakly excites
  12. L27:1358 weakly excites
  13. L21:2104 weakly suppresses
  14. L23:1199 weakly excites
  15. L22:2204 weakly suppresses
  16. L25:618 weakly suppresses
  17. L22:3016 weakly suppresses
  18. L23:2768 weakly excites
  19. L27:881 weakly excites "as a result" before "a"
  20. L22:717 weakly excites
  21. L26:235 weakly excites
  22. L23:1592 weakly excites
  23. L25:2921 weakly suppresses
  24. L23:2644 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 -8.34 and -7.971 values between -7.23 and -6.871 values between -6.87 and -6.503 values between -6.50 and -6.131 values between -6.13 and -5.773 values between -5.40 and -5.031 values between -5.03 and -4.664 values between -4.66 and -4.302 values between -4.30 and -3.931 values between -3.93 and -3.565 values between -3.56 and -3.205 values between -3.20 and -2.832 values between -2.83 and -2.464 values between -2.46 and -2.096 values between -2.09 and -1.7313 values between -1.73 and -1.3621 values between -1.36 and -0.99162 values between -0.99 and -0.63717 values between -0.63 and -0.262050 values between -0.26 and 0.11606 values between 0.11 and 0.48141 values between 0.48 and 0.8472 values between 0.84 and 1.2138 values between 1.21 and 1.5816 values between 1.58 and 1.959 values between 1.95 and 2.316 values between 2.31 and 2.685 values between 2.68 and 3.053 values between 3.05 and 3.413 values between 3.41 and 3.782 values between 4.15 and 4.522 values between 4.52 and 4.881 values between 5.25 and 5.621 values between 5.98 and 6.35offnegativepositiveoften

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. hài
  2. único
  3. tarde
  4. ActionCode
  5. عادة
  6. 光彩
  7. 牢
  8. wä
  9. geries
  10. VAS

Suppresses

  1. planning
  2. 規劃
  3. 認識
  4. Planning
  5. Planning
  6. 立志
  7. 准备
  8. 发现问题
  9. 準備
  10. 认识