Neuron Explorer

Qwen3-0.6B

"and" in noun lists, "through" in ranges

A neuron in layer 1 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 "and" at the end of a list of nouns, such as "researchers, and government officials". Its strongest negative reaction is on the word "through" in ranges like "Sections 1 through 8". The steering outputs did not clearly reflect either pattern.

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

    ulture.⏎Now in its ninth year, the event drew 40 per cent of its attendees from the seafood business. Industry personnel joined conservationists, researchers, and government officials in discussing a range of sustainability issues - marine spatial planning, traceability

  2. #2 peak

    Problem of Childhood Obesity Within a Generation. The task force, part of the First Lady’s Let’s Move campaign, offers goals, benchmarks, and measureable outcomes that will help the country tackle the childhood obesity epidemic one child,

  3. #3 peak

    . The topography, rainfall and water supply were different from that of the East. The Easterners had a forest background. The traditions, laws, weapons, and the general approach to current problems were different. In this environment the influence of ge

  4. #4 peak

    56 caregiving failures) from 1985 to 2003. The fatality data was gathered from media reports, legal cases, and state records. Three forms of child care were investigated: child care centers, n

  5. #5 peak

    his international fame as<|endoftext|>“Paradise Found and Lost” from Daniel J. Boorstin’s The Discoverers, embodies Columbus’ emotions, ideas, and hopes. Boorstin, a former Librarian of Congress, leads the

  6. #6 peak

    actual picture of the Biggest Beaver Dam in the World:<|endoftext|>Editorial and Revision Process⏎Preparing an Inquiry article for publication is a collaboration between authors, editors, and faculty mentors. The authors of Inquiry articles often begin with a report or essay describing

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 been thinking about lately, and I have been thinking about lately. So I will do my part of the task. Okay, I have been thinking about lately, and I have been thinking about lately. So I will do my part

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

    29 CFR 1910.1200. The SDS preparers may also include additional information in various section(s).⏎Sections 1 through 8 contain general information about the chemical, identification, hazards, composition, safe

  2. #2 peak

    programs. Posters and interpretive texts provide additional information, context and commentary.⏎Exhibit panels are organized in sections falling into two major groups. Sections 1 through 7 tell the story of agriculture and farmworkers in our area by decade over

  3. #3 peak

    8 and 10 students take a writing test which consists of an essay and multiple choice questions. A reading and mathematics test is given in grades 3 through 10. In grades 5, 8

  4. #4 peak

    , summarizing the essential findings. The SRREN consists of 11 chapters. Chapter 1 sets the context for RE and climate change; Chapters 2 through 7 provide information on six RE technologies, and Chapters 8 through 1

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. L0:1224 suppresses
  2. L0:1301 suppresses
  3. L0:178 suppresses
  4. L0:605 weakly suppresses 'of' after an action noun
  5. L0:2426 weakly suppresses
  6. L0:1 weakly suppresses
  7. L0:493 weakly suppresses
  8. L0:1888 weakly excites
  9. L0:1523 weakly suppresses
  10. L0:993 weakly excites
  11. L0:1866 weakly excites
  12. L0:1037 weakly excites
  13. L0:582 weakly suppresses
  14. L0:451 weakly suppresses
  15. L0:1454 weakly excites
  16. L0:853 weakly excites
  17. L0:1513 weakly suppresses
  18. L0:2450 weakly excites
  19. L0:1309 weakly excites
  20. L0:647 weakly excites
  21. L0:2571 weakly suppresses
  22. L0:1343 weakly suppresses
  23. L0:525 weakly excites
  24. L0:2227 weakly suppresses

Feeds later layers

  1. L17:3031 suppresses
  2. L11:527 excites
  3. L2:1684 weakly excites
  4. L2:2302 weakly suppresses
  5. L13:656 weakly suppresses 'and' or comma between paired items
  6. L4:385 weakly excites multiplication and scaling in math
  7. L11:1325 weakly suppresses
  8. L22:306 weakly excites
  9. L23:716 weakly suppresses
  10. L10:1967 weakly excites
  11. L24:2760 weakly excites
  12. L2:2654 weakly excites
  13. L13:35 weakly suppresses
  14. L3:1333 weakly excites
  15. L5:46 weakly excites
  16. L7:2121 weakly suppresses
  17. L10:2938 weakly excites
  18. L4:627 weakly excites
  19. L4:457 weakly excites
  20. L7:2458 weakly suppresses
  21. L10:2118 weakly excites
  22. L18:1091 weakly excites
  23. L8:841 weakly excites
  24. L8:2371 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.

2 values between -1.89 and -1.811 values between -1.73 and -1.661 values between -1.66 and -1.581 values between -1.34 and -1.261 values between -1.26 and -1.181 values between -1.10 and -1.021 values between -0.94 and -0.863 values between -0.86 and -0.791 values between -0.79 and -0.711 values between -0.71 and -0.632 values between -0.63 and -0.555 values between -0.55 and -0.475 values between -0.47 and -0.398 values between -0.39 and -0.3130 values between -0.31 and -0.2391 values between -0.23 and -0.15448 values between -0.15 and -0.071453 values between -0.07 and 0.011179 values between 0.01 and 0.09440 values between 0.09 and 0.17136 values between 0.17 and 0.2448 values between 0.24 and 0.3213 values between 0.32 and 0.408 values between 0.40 and 0.486 values between 0.48 and 0.565 values between 0.56 and 0.643 values between 0.64 and 0.725 values between 0.72 and 0.801 values between 0.80 and 0.885 values between 0.88 and 0.962 values between 0.96 and 1.041 values between 1.04 and 1.111 values between 1.19 and 1.27offnegativepositiveoften

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. ="⏎
  2. .Raw
  3. Natural
  4. jr
  5. .jasper
  6. =__
  7. andes
  8. cost
  9. icals
  10. иков

Suppresses

  1. نا
  2. ourselves
  3. haft
  4. 息
  5. etCode
  6. elem
  7. ""),
  8. ATE
  9. Explore
  10. elem