Neuron Explorer

Qwen3-0.6B

weak, mixed: verbs and adverbs in claims

A neuron in layer 16 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 weakly and inconsistently, mostly on verbs and helper words in sentences that make claims about health or evidence, such as "are interconnected" or "those extraordinary claims." It turns negative on adverbs like "deftly" and "elegantly." No clear pattern came through when the model was steered.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    to our overall physical wellbeing – recent federal government health funding gives recognition to this. It might seem obvious, but we often forget that the health of mind and body are interconnected, and when one functions well, the other improves, too.⏎A number

  2. #2 peak

    from traumatized males into fertilized wild-type oocytes reproduced the behavioral and metabolic alterations in the resulting offspring.”⏎Unfortunately, the paper provides no evidence to back up those extraordinary claims. It is, regrettably, a prime example of over-f

  3. #3 peak

    of your overall competence. When someone makes an inattentive mistake, it’s doesn’t require much effort simply to assume that they don’t know what they’re doing.⏎But of course, ADHDers are much more than their inattent

  4. #4 peak

    . Frank Pellegrini, DVM, has shown that colonic ulcers affect perhaps 60 percent of performance horses.⏎Fortunately, ulcers can be treated once they are diagnosed, and affected horses can continue their careers

  5. #5 peak

    plastics. However while the risk of exposure via exercise products would appear to be low, users such as children or pregnant women should avoid using products containing phthalates until more research is completed.⏎How People Are Exposed to Phthalates⏎People

  6. #6 peak

    yard, inventory your yard and work with the features you have. With the aid of field guides, you may locate hidden treasures in the form of wildflowers you were unaware of. Perhaps a stump, fallen log, or large

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 a, I have a, and I have a, I have a, and I have a, and I have a, and I have a, and I have a, and I have a, and I have a, 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

    Fragility: Why It’s So Hard for White People to Talk About Racism by Robin Diangelo⏎“Antiracist educator Robin DiAngelo deftly illuminates the phenomenon of white fragility and “allows us to understand racism as

  2. #2 peak

    always saying things like “animation makes our websites (or presentations, or whatever) come alive.” It might be overused, but it’s a phrase that elegantly captures the purpose of animation in web design.⏎That illusion of motion, when applied

  3. #3 peak

    recommendations from the American College of Obstetricians and Gynecologists (ACOG) are quite clear on that issue.⏎In those same recommendations, ACOG also states

  4. #4 peak

    ever been found anywhere in the world despite the most diligent and prolonged search in the fossil record.⏎In their attempts to gather evidence for evolution, evolutionists have unwittingly proven by their own efforts that evolution cannot have happened at all!⏎The person who

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. L8:94 excites
  2. L13:685 suppresses
  3. L14:2568 excites
  4. L14:659 excites
  5. L6:439 suppresses
  6. L6:1880 excites "by" introducing a deadline or means
  7. L14:1158 excites 'to' starting a purpose clause
  8. L13:1161 suppresses
  9. L14:1133 suppresses unclear: commas and nouns before an aside
  10. L7:681 excites 'from' after prevent or stop
  11. L15:1176 excites
  12. L9:226 suppresses 'to' starting an infinitive
  13. L12:1000 excites
  14. L6:2359 excites
  15. L13:909 excites 'to' before a verb after a noun
  16. L10:616 excites
  17. L13:2681 excites
  18. L15:448 excites
  19. L6:1133 excites "to" before an infinitive verb
  20. L8:723 suppresses
  21. L6:2057 suppresses
  22. L6:914 excites
  23. L14:1887 excites
  24. L15:32 excites

Feeds later layers

  1. L18:2678 excites
  2. L19:617 weakly suppresses
  3. L17:2213 weakly excites
  4. L18:2572 weakly excites
  5. L17:2659 weakly excites
  6. L20:16 weakly suppresses
  7. L18:1852 weakly excites
  8. L20:853 weakly excites
  9. L17:2953 weakly excites
  10. L18:2929 weakly excites comma after a finished clause
  11. L17:58 weakly suppresses
  12. L22:337 weakly excites
  13. L20:1618 weakly excites 'be' after 'to' or 'need to'
  14. L20:2418 weakly excites
  15. L17:893 weakly excites
  16. L27:443 weakly excites
  17. L21:123 weakly suppresses
  18. L25:860 weakly excites
  19. L19:761 weakly excites
  20. L21:2320 weakly excites
  21. L27:1171 weakly excites
  22. L19:1166 weakly excites
  23. L27:2609 weakly excites
  24. L17:790 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 -8.98 and -8.661 values between -7.06 and -6.742 values between -6.74 and -6.421 values between -6.42 and -6.091 values between -6.09 and -5.776 values between -5.45 and -5.136 values between -5.13 and -4.8110 values between -4.81 and -4.493 values between -4.49 and -4.177 values between -4.17 and -3.859 values between -3.85 and -3.5314 values between -3.53 and -3.2015 values between -3.20 and -2.8817 values between -2.88 and -2.5621 values between -2.56 and -2.2418 values between -2.24 and -1.9226 values between -1.92 and -1.6038 values between -1.60 and -1.2838 values between -1.28 and -0.9662 values between -0.96 and -0.64168 values between -0.64 and -0.311085 values between -0.31 and 0.011110 values between 0.01 and 0.33572 values between 0.33 and 0.65321 values between 0.65 and 0.97158 values between 0.97 and 1.2987 values between 1.29 and 1.6160 values between 1.61 and 1.9322 values between 1.93 and 2.2615 values between 2.26 and 2.5810 values between 2.58 and 2.902 values between 2.90 and 3.222 values between 3.54 and 3.86offnegativepositiveoften

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. anse
  2. をしている
  3. estar
  4. itzer
  5. :convert
  6. .GetService
  7. Make
  8. 新形势下
  9. ﹍
  10. 之家

Suppresses

  1. flock
  2. 一
  3. inated
  4. dial
  5. ũ
  6. vak
  7. immediately
  8. either
  9. .PropertyType
  10. ительно