Neuron Explorer

Qwen3-0.6B

'a' before a noun phrase

A neuron in layer 2 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 small word 'a' when it comes after a preposition or verb, such as 'a variety of', 'a changing world' or 'a nation'. It is mostly quiet on other words. Turning it up made the model's text drift into unrelated topics, so the steering results don't show a clear effect.

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

    into the effects of gut microbiota on mosquito life traits. Metagenomic RNA-Seq has become an important tool to analyze transcriptomes from various microbes present in a microbial community. Messenger RNA usually comprises only 1-3% of total RNA

  2. #2 peak

    decrease in the art appreciation and there have been discussions concerning its future.⏎Today, lots of people dispute whether art still has an essential duty in the social background of a nation or society. Numerous experts agree that art and also the creation of artworks is

  3. #3 peak

    truly sustainable strategy tries to understand the impact of building, or of not building, and to address the environmental, ecological, cultural, economic, and financial issues of a place.⏎Even after differentiating the terms, using ‘green’ and ‘s

  4. #4 peak

    statement under the charter evolved to be the most widely accepted among professionals in the field. The statement heavily encourages environmental education to be a lifelong journey focused on responding to a changing world. It suggests environmental education should provide “the provision of skills and attributes

  5. #5 peak

    Nasser al-Missned, chair of the foundation, who said, “Today, we place great faith in the power of education to prepare world citizens for a peaceful and cooperative future and to prepare citizens of our individual nations for the cultural transformations

  6. #6 peak

    , animal cruelty laws have been in place since 1879. While earlier versions of the laws dealt with cattle and livestock, today Texas cruelty laws protect a variety of animals from inhumane treatment. These laws define which actions are “

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 writing about my own life, and I have been thinking about how I feel like I have been living in the city, and I have been writing about my own experience, and I have been writing about how I feel like I

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

    as TBI, is thought to be extremely difficult to recover from. Well how about we inject some “female hormones” and see what happens?⏎In a recent article, Rita Rubin writes of an individual who suffered a serious injury from a traffic incident

  2. #2 peak

    gut reaction might be to assume that diet soda is a better option. Unfortunately, research has also found a connection between diet soda and brain health.⏎One study published in Stroke found that participants who had consumed one diet soda daily were almost three times more

  3. #3 peak

    the truth of the matter is that humans aren't built with the skills in place to incorporate all this new technology readily without some stress."⏎In a New York Times article last year, one writer documented her own experience with FOMO.⏎

  4. #4 peak

    -based care.⏎But such rapid digitalization comes at a cost. The healthcare industry today is the most targeted industry in terms of cyber security threats. A recent study by Critical insights (A cyber security company) shows that in 2021

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. L1:237 excites
  2. L1:376 weakly suppresses
  3. L1:1856 weakly excites
  4. L0:1830 weakly excites
  5. L1:2229 weakly suppresses
  6. L1:334 weakly suppresses
  7. L0:370 weakly excites
  8. L0:523 weakly excites
  9. L0:2723 weakly suppresses
  10. L0:1320 weakly suppresses
  11. L0:1224 weakly suppresses
  12. L1:1722 weakly excites
  13. L1:483 weakly excites
  14. L1:2801 weakly excites
  15. L0:2243 weakly excites
  16. L1:790 weakly excites
  17. L1:2513 weakly excites
  18. L0:2039 weakly excites
  19. L0:3014 weakly excites
  20. L0:178 weakly suppresses
  21. L0:2636 weakly suppresses
  22. L1:1923 weakly excites
  23. L0:2258 weakly suppresses
  24. L1:1150 weakly suppresses

Feeds later layers

  1. L3:618 suppresses
  2. L4:1454 weakly suppresses
  3. L11:1507 weakly excites
  4. L7:493 weakly excites 'this' pointing back to a topic
  5. L7:2571 weakly suppresses
  6. L13:1562 weakly suppresses unclear: words that set up a qualifier
  7. L8:203 weakly excites
  8. L6:452 weakly suppresses
  9. L4:218 weakly excites
  10. L4:1216 weakly suppresses
  11. L23:892 weakly suppresses "the" in "over the past" time spans
  12. L7:2411 weakly suppresses
  13. L7:2932 weakly suppresses
  14. L23:34 weakly suppresses
  15. L9:2 weakly excites
  16. L4:923 weakly suppresses
  17. L19:1898 weakly excites
  18. L4:340 weakly excites
  19. L9:1496 weakly excites
  20. L8:1377 weakly excites
  21. L19:1836 weakly suppresses
  22. L3:125 weakly excites verb right after a subject in clauses
  23. L5:1279 weakly suppresses
  24. L12:1991 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 -2.03 and -1.941 values between -1.78 and -1.701 values between -1.53 and -1.451 values between -1.36 and -1.281 values between -1.28 and -1.201 values between -1.12 and -1.032 values between -1.03 and -0.953 values between -0.95 and -0.873 values between -0.87 and -0.784 values between -0.78 and -0.704 values between -0.70 and -0.628 values between -0.62 and -0.5312 values between -0.53 and -0.4528 values between -0.45 and -0.3742 values between -0.37 and -0.29142 values between -0.29 and -0.20357 values between -0.20 and -0.12836 values between -0.12 and -0.041580 values between -0.04 and 0.05521 values between 0.05 and 0.13172 values between 0.13 and 0.2174 values between 0.21 and 0.3040 values between 0.30 and 0.3821 values between 0.38 and 0.4613 values between 0.46 and 0.5513 values between 0.55 and 0.636 values between 0.63 and 0.714 values between 0.71 and 0.7910 values between 0.79 and 0.881 values between 0.88 and 0.961 values between 1.04 and 1.133 values between 1.13 and 1.212 values between 1.21 and 1.29offnegativepositiveoften

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. 发展目标
  2. \Foundation
  3. mín
  4. intree
  5. eton
  6. Erect
  7. +')
  8. Veget
  9. Riders
  10. confort

Suppresses

  1. 同期
  2. colleague
  3. entry
  4. ah
  5. collabor
  6. 出
  7. ynı
  8. shi
  9. isser
  10. qu