Neuron Explorer

Qwen3-0.6B

healthcare and service phrases

A neuron in layer 9 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the word in a phrase where the next word is likely to be 'services', 'care', 'access' or 'quality', especially in health and cloud computing text. Its strongest examples are all about health care, service or access. Turning it up did not make the model talk about health, so the link is only seen in the text it reads.

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

    Cloud Computing?⏎In simple terms, cloud computing is a technology used for storing and accessing data, application and programs over the Internet. Cloud computing refers to the delivery of hosted and computing services including servers, storage, databases, networking, analytics, and

  2. #2 peak

    edition of Health at a Glance: Europe presents a set of key indicators of health status, determinants of health, health care resources and activities, quality of care, health expenditure and financing in 35 European countries, including the 2

  3. #3 peak

    were run out of the area. While growing up there it never occurred to me<|endoftext|>In the United States, medical insurance plays a crucial role in ensuring access to quality healthcare. With a complex healthcare system, navigating the intricacies of medical insurance can

  4. #4 peak

    aspect of reducing health disparities is moving health care providers, staff, administrators, and practices toward increased cultural competence and proficiency. Effective delivery of culturally and linguistically appropriate service in cross-cultural settings is identified as cultural competence

  5. #5 peak

    intervention for cervical cancer screening for African American and Native American women.⏎A Roadmap and Best Practices for Organizations to Reduce Racial and Ethnic Disparities in Health Care⏎- 1. Interventions to Improve Decision Making and Reduce Racial

  6. #6 peak

    00,000 of whom were living in sub-Saharan Africa. Though life-saving antiretroviral treatment is available, millions of people still cannot access it: just 32% of children who are HIV-positive are currently on

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 1960s, 1970s, and 2000s. Please explain what the key points are for the 1960s, 1970s, 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

    ly described by the Monty Python comedy troupe as “taxing foreigners living abroad.” But the fact that

  2. #2 peak

    12 HealthDay. All rights reserved. URL:http://www.healthscout.com/template.asp?id=6695

  3. #3 peak

    productivity of teams, but it can also potentially hurt relationships with customers or damage the reputation of the business.⏎How dirty data happens⏎In Salesforce and most other CRMs, the origin of dirty data can usually be pinpointed by one or more of

  4. #4 peak

    oil analysis—also can tell the condition of equipment, which allows us to respond to potential failures at an early stage.⏎In short, the proper use of oil analysis can help:⏎- Improve

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. L6:2977 suppresses
  2. L5:2579 excites
  3. L6:2955 suppresses
  4. L6:1384 weakly suppresses
  5. L3:2739 weakly suppresses
  6. L8:2498 weakly excites
  7. L3:1136 weakly excites
  8. L4:1397 weakly suppresses
  9. L6:2029 weakly suppresses row and column words in grids and tables
  10. L7:3016 weakly excites
  11. L1:2551 weakly excites
  12. L5:2268 weakly suppresses
  13. L1:2683 weakly excites
  14. L6:2130 weakly excites
  15. L7:2100 weakly excites
  16. L7:1553 weakly excites
  17. L6:2005 weakly suppresses
  18. L8:1421 weakly excites
  19. L6:1091 weakly suppresses
  20. L2:2510 weakly suppresses
  21. L4:2797 weakly suppresses
  22. L1:1942 weakly excites
  23. L6:825 weakly suppresses
  24. L8:2577 weakly suppresses

Feeds later layers

  1. L22:2087 excites
  2. L12:1476 excites
  3. L20:1654 excites
  4. L24:1177 weakly suppresses
  5. L20:1063 weakly excites
  6. L11:1113 weakly excites
  7. L10:847 weakly suppresses
  8. L19:2203 weakly excites
  9. L20:2649 weakly suppresses
  10. L24:1980 weakly excites
  11. L17:3062 weakly excites
  12. L11:727 weakly excites
  13. L10:2147 weakly excites
  14. L22:2340 weakly excites
  15. L19:2154 weakly suppresses
  16. L11:1628 weakly suppresses
  17. L20:801 weakly suppresses
  18. L17:1162 weakly excites
  19. L17:2615 weakly suppresses
  20. L13:3048 weakly excites
  21. L11:1000 weakly excites
  22. L23:1101 weakly excites
  23. L22:2860 weakly excites
  24. L12:420 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.

5 values between -0.73 and -0.5843 values between -0.58 and -0.43262 values between -0.43 and -0.281011 values between -0.28 and -0.131733 values between -0.13 and 0.02638 values between 0.02 and 0.17129 values between 0.17 and 0.3228 values between 0.32 and 0.4715 values between 0.47 and 0.6214 values between 0.62 and 0.779 values between 0.77 and 0.933 values between 0.93 and 1.086 values between 1.08 and 1.232 values between 1.23 and 1.384 values between 1.38 and 1.531 values between 1.53 and 1.681 values between 1.68 and 1.831 values between 1.83 and 1.981 values between 1.98 and 2.131 values between 2.13 and 2.281 values between 5.14 and 5.29offnegativepositiveoften

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. services
  2. 服务
  3. service
  4. -service
  5. .services
  6. services
  7. 服务能力
  8. 医疗服务
  9. _services
  10. сервис

Suppresses

  1. 这首歌
  2. 独角
  3. Pose
  4. 唧
  5. storybook
  6. estone
  7. 赛车
  8. 蛴
  9. 包袱
  10. 方向盘