Neuron Explorer

Qwen3-0.6B

comma or verb after a researcher or study subject

A neuron in layer 13 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 the word right after the name of a study, team or researcher in a news story, such as the comma in "...Purdue University professor of geophysics," that leads into what they did. Turned up, the text did not clearly reflect this, so the link to the output is weak.

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

    the Earth's landscape.⏎An international research team led by Eric Calais, a Purdue University professor of geophysics, was able to measure ground displacements as two tectonic plates in Africa moved apart

  2. #2 peak

    We know more about where the Milky Way’s spiral arms are.”⏎Named GLIMPSE360 (Galactic Legacy Mid-Plane Survey Extraordinaire project), the deep infrared survey captures only about 3% of the sky, but because

  3. #3 peak

    age phase.⏎A research carried out by the Department of Agricultural and Environmental Sciences - Production, Landscape, Agroenergy of the Università degli Studi di Milano aims at providing an innovative remuage method which is alternative to available solutions. The

  4. #4 peak

    few were taking the study drugs as directed.⏎The trial, known as the Vaginal and Oral Interventions to Control the Epidemic (VOICE) study, was designed to evaluate the safety and efficacy of three HIV prevention strategies compared to placebo

  5. #5 peak

    calories than that, though not all from sugar.⏎The Californian scientists who conducted the 175-nation study, published this week in PLOS ONE, showed that it is not merely the amount of calories in

  6. #6 peak

    . Proceedings of the National Academy of Sciences USA (PNAS) has published the study online this week.⏎Argentinian scientist Andrés Dekanty, a Juan de la Cierva researcher in the Milán lab and first

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: What I have done? What's the point of the thing? What's the point of the? What's the point? what? ? ? ? ? ? ? ? ? ??

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

    cent lower risk of developing Alzheimer's than those with scores in the lowest 10 per cent.⏎The researchers, writing in the journal Archives of General Psychiatry, also found that conscientiousness was associated with a slower rate of mental decline.⏎But

  2. #2 peak

    response variability). Moreover, people with high ADHD-traits were more vulnerable to sleep deprivation and showed greater impairment than those with low ADHD-traits.⏎The effects were also related to the most prominent type of subclinical ADHD-like symptom, in that after

  3. #3 peak

    a virus that infects bacteria can help rather than harm the bacteria's animal host, said Oliver, who⏎earned his doctorate at the UA.The researchers also tested strains of aphids whose bacteria had once been⏎The virus, known as

  4. #4 peak

    the skin after an injury to inhibit nerve signals and ease pain in a way similar to lidocaine.⏎In the current study, scientists found that the chemical is also very active in deeper tissues affected by acupuncture. The Rochester researchers looked at the effects

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. L12:978 strongly excites
  2. L11:1391 suppresses
  3. L7:741 excites
  4. L10:2988 excites
  5. L7:1053 suppresses
  6. L7:1766 excites
  7. L11:880 suppresses
  8. L7:33 excites
  9. L12:596 suppresses
  10. L10:1663 suppresses
  11. L11:2606 suppresses
  12. L11:665 excites
  13. L8:1775 excites
  14. L10:72 excites
  15. L11:2100 excites comma before a describing clause
  16. L12:1051 excites
  17. L7:1454 weakly suppresses
  18. L8:169 weakly suppresses
  19. L12:2366 weakly excites
  20. L11:1974 weakly excites
  21. L10:1885 weakly excites
  22. L12:76 weakly suppresses
  23. L0:2900 weakly excites
  24. L8:960 weakly excites

Feeds later layers

  1. L14:2224 strongly suppresses web citations, urls and dateline dashes
  2. L14:1077 strongly suppresses
  3. L15:1154 suppresses
  4. L18:796 suppresses
  5. L14:994 excites
  6. L14:2434 excites
  7. L14:455 excites
  8. L17:192 suppresses
  9. L14:1668 excites
  10. L14:2806 suppresses
  11. L15:2845 excites
  12. L14:681 excites
  13. L14:2370 excites
  14. L14:1317 suppresses
  15. L17:2281 excites
  16. L14:883 suppresses
  17. L15:2134 suppresses
  18. L14:2826 excites
  19. L15:1435 weakly excites
  20. L15:1478 weakly excites
  21. L16:2832 weakly excites
  22. L17:479 weakly excites
  23. L16:1904 weakly excites
  24. L15:1687 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.

4 values between -1.61 and -1.382 values between -1.38 and -1.167 values between -1.16 and -0.9324 values between -0.93 and -0.7153 values between -0.71 and -0.48161 values between -0.48 and -0.26826 values between -0.26 and -0.031677 values between -0.03 and 0.19429 values between 0.19 and 0.42271 values between 0.42 and 0.64154 values between 0.64 and 0.87102 values between 0.87 and 1.0971 values between 1.09 and 1.3145 values between 1.31 and 1.5423 values between 1.54 and 1.7618 values between 1.76 and 1.9914 values between 1.99 and 2.213 values between 2.21 and 2.447 values between 2.44 and 2.663 values between 2.66 and 2.893 values between 2.89 and 3.111 values between 3.11 and 3.345 values between 3.34 and 3.562 values between 4.01 and 4.241 values between 5.14 and 5.361 values between 5.36 and 5.591 values between 7.16 and 7.38offnegativepositiveoften

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. THEN
  2. ده
  3. &);⏎⏎
  4. ukkit
  5. .react
  6. parameter
  7. _then
  8. 則
  9. lik
  10. itarian

Suppresses

  1. 说起
  2. venge
  3. enate
  4. 另一种
  5. 毁灭
  6. LC
  7. enus
  8. 展览会
  9. üns
  10. 谈起