Neuron Explorer

Qwen3-0.6B

opening parenthesis after a name

A neuron in layer 0 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 reacts to an opening bracket that comes right after a name or acronym, such as 'Hawaii Wildlife Fund (HWF)' or 'Copyright (c)'. Turning it up did not clearly change the written text.

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

    <|endoftext|>Charles Messier (France) had just completed an observation of Jupiter and was in the process of observing

  2. #2 peak

    Hawaii Wildlife Fund (HWF) collected and hauled away 15,000 pounds

  3. #3 peak

    it is more comfortable to wear it over a shirt than as a single⏎piece of clothing.⏎Q4. What do you wear on veshti?⏎<|endoftext|>Copyright (c) Arvin S. Quist⏎INTRODUCTION TO CLASSIFICATION⏎THE

  4. #4 peak

    of creating a “thermostatic mechanism” that would counter-balance the dominant presence of electronic media in the lives of children.⏎How to Cite⏎Copyright (c) 2017 Robert Albrecht⏎Copyrights and publishing

  5. #5 peak

    library in the app or on the desktop website.<|endoftext|>The ultimate guide to Web animation | Webdesigner Depot⏎Animation is one of humanity’s life-long dreams (if some historians are to be believed). The theory begins with cave paintings: In

  6. #6 peak

    Children’s games are significant;⏎They have a tremendous role shaping the child’s life”⏎– e² Young Engineers –⏎Our most popular program introduces elementary school children (grades 1- 5) to the basic principles of mathematics, science and

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: this is the time to write about the new content of the story. So here is the story. I have been thinking about lately: this is the time to write about the new content of the story. So here is the story. 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

    of the played cards error-free.⏎Novices translate simple words, while advanced players translate more difficult words. In addition to vocabulary, there are also sentences and idiomatic expressions. New Amigos also includes geographical information and cards focused on culture, business

  2. #2 peak

    be written on this subject. I'm guessing the "multi-million line production code base" in question is the Dropbox code base, and maybe Dropbox has an idiomatic way of writing Python with lots of "isinstance()"s.⏎> In general

  3. #3 peak

    vic acids, increased ketosis,⏎and hypoglycemia. Treatment consists of continuous glucose supplementation⏎to avoid gluconeogenesis. Hereditary fructose intolerance presents with⏎vomiting after the introduction of fruit juice or table sugar

  4. #4 peak

    of carbohydrates occurs in fructose-1-6⏎diphosphate deficiency and in fructose-1-phosphate aldolase deficiency⏎(hereditary fructose intolerance). Fructose-1-6 diphosphate deficiency⏎produces accumulation

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

None: this is the first layer.

Feeds later layers

  1. L9:602 weakly suppresses
  2. L17:1509 weakly excites
  3. L10:133 weakly excites
  4. L22:2136 weakly excites
  5. L13:2447 weakly excites
  6. L14:90 weakly excites
  7. L12:878 weakly excites
  8. L3:1774 weakly excites
  9. L8:2379 weakly excites
  10. L9:734 weakly excites
  11. L16:1041 weakly suppresses
  12. L10:842 weakly suppresses
  13. L17:775 weakly excites
  14. L14:370 weakly excites
  15. L3:2300 weakly suppresses
  16. L23:90 weakly excites
  17. L11:3030 weakly excites
  18. L23:1149 weakly excites
  19. L11:1703 weakly suppresses
  20. L16:2190 weakly excites
  21. L20:2695 weakly suppresses
  22. L16:264 weakly excites opening parenthesis in citations and asides
  23. L1:587 weakly suppresses
  24. L18:2320 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.

3 values between -0.68 and -0.6016 values between -0.35 and -0.2724 values between -0.27 and -0.19168 values between -0.19 and -0.10997 values between -0.10 and -0.022362 values between -0.02 and 0.06281 values between 0.06 and 0.1434 values between 0.14 and 0.235 values between 0.23 and 0.314 values between 0.31 and 0.391 values between 0.64 and 0.721 values between 1.13 and 1.211 values between 1.29 and 1.371 values between 1.37 and 1.461 values between 1.54 and 1.621 values between 1.70 and 1.791 values between 1.79 and 1.872 values between 1.87 and 1.952 values between 2.03 and 2.112 values between 2.28 and 2.361 values between 2.52 and 2.61offnegativepositiveoften

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. ší
  2. symbol
  3. arriv
  4. iş
  5. ała
  6. ournal
  7. .ERR
  8. możliwe
  9. orro
  10. 跑了

Suppresses

  1. toJson
  2. plementary
  3. plement
  4. phas
  5. peria
  6. 按照
  7. 前者
  8. lius
  9. -fed
  10. rescia