Neuron Explorer

Qwen3-0.6B

'the' starting an object phrase

A neuron in layer 17 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

The neuron lights up on the word 'the' when it comes right after a verb or preposition and starts the thing being acted on, like 'consumed the glucose' or 'evaporate the water'. Turning it up made the model's writing drift into rambling, off-topic sentences, so the steering effect is not very readable.

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

    after generation, for twenty years (he deserves some marks for persistence!). The culture medium had a little glucose but lots more citrate, so once the microbes consumed the glucose, they would continue to grow only if they could evolve some way of using

  2. #2 peak

    an apple orchard. York’s unit was feverishly digging foxholes as the men heard the shells edging closer. “And then bang!—one of the big shells struck the ground right in front of us and we all went up in

  3. #3 peak

    ure_ is indeed an old word, but, I think, always applied to things, never to persons.⏎Maybe get a guard dog as well to keep out the money grubbing terrorists!. .there are just too many perfectly good homes going to

  4. #4 peak

    inator. This will filter the mud from the water, which will then flow into the desalinator. There will be a heater within the base to evaporate the water, leaving the salt on the bottom. The water will condense and flow

  5. #5 peak

    use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.<|endoftext|>- Major Systems >⏎- The Plumbing Rough-In

  6. #6 peak

    Spread seed with a drop, broadcast, or handheld spreader.⏎- At low seeding rates, sawdust, sand, or other suitable material mixed⏎with the seed aids in obtaining uniform coverage.⏎- Cover seed to a depth of 1

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: the number of the month of January, which is the first, and it is first; it is, as you might, seen, by the; and, in the same; of; and; of;; -; -; -

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

    JSTOR JSTOR is part of ITHAKA, a not-for-profit organization helping the academic community use digital technologies to preserve the scholarly record and to advance research and teaching in sustainable ways ... Types of Sources for

  2. #2 peak

    culture to think of heat as risky,” said Bernstein, a pediatrician at Boston Children’s Hospital and co-director of the Center for Climate, Health, and the Global Environment at Harvard’s School of Public Health.⏎As global temperatures rise and extreme

  3. #3 peak

    supply chains over the edge.⏎Water is an unseen, vital input for all economic activity. Beyond agriculture, water is critical for power generation, mining, industry and the consumer products we rely on every single day. Like the estimated 300

  4. #4 peak

    human health. By supporting and applying epigenetics research on a national scale, CEEHRC is harnessing its great potential for major breakthroughs in understanding the origins of health and also the etiology, treatment and prevention of many of the

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. L10:120 strongly suppresses
  2. L9:662 strongly suppresses
  3. L10:1377 strongly suppresses
  4. L11:1168 strongly excites
  5. L4:73 strongly suppresses
  6. L15:888 strongly excites
  7. L16:792 strongly excites "the" inside a prepositional phrase
  8. L16:652 suppresses
  9. L16:419 excites
  10. L9:2198 excites
  11. L16:1594 excites
  12. L9:24 excites "the" before a named organisation or title
  13. L13:1234 suppresses
  14. L11:875 suppresses
  15. L15:1888 excites
  16. L2:205 excites
  17. L10:1644 excites
  18. L16:2248 suppresses
  19. L16:2062 excites "other" and contrast words
  20. L10:1162 excites
  21. L10:321 suppresses
  22. L15:1848 suppresses
  23. L15:2253 excites
  24. L3:344 excites

Feeds later layers

  1. L18:641 suppresses
  2. L18:799 suppresses
  3. L20:932 weakly suppresses 'of' after a count or quantity
  4. L19:3016 weakly suppresses
  5. L19:1744 weakly suppresses 'the' in mid-sentence noun phrases
  6. L18:2411 weakly suppresses
  7. L18:745 weakly excites
  8. L18:1782 weakly suppresses
  9. L19:2982 weakly excites
  10. L18:2370 weakly suppresses
  11. L19:2208 weakly suppresses
  12. L24:1410 weakly excites
  13. L19:2119 weakly suppresses
  14. L18:134 weakly suppresses
  15. L23:1966 weakly suppresses
  16. L23:1235 weakly excites
  17. L18:2651 weakly suppresses
  18. L20:2238 weakly suppresses 'the' in 'on the other hand' phrases
  19. L24:957 weakly excites
  20. L23:2578 weakly suppresses
  21. L19:566 weakly excites
  22. L19:2309 weakly suppresses
  23. L19:2356 weakly excites
  24. L23:24 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 -15.33 and -14.821 values between -12.78 and -12.271 values between -11.76 and -11.251 values between -11.25 and -10.741 values between -10.74 and -10.231 values between -10.23 and -9.732 values between -9.22 and -8.713 values between -8.71 and -8.203 values between -8.20 and -7.691 values between -7.69 and -7.183 values between -7.18 and -6.671 values between -6.67 and -6.164 values between -6.16 and -5.656 values between -5.65 and -5.145 values between -5.14 and -4.6310 values between -4.63 and -4.1210 values between -4.12 and -3.614 values between -3.61 and -3.1011 values between -3.10 and -2.5917 values between -2.59 and -2.0821 values between -2.08 and -1.5832 values between -1.58 and -1.0754 values between -1.07 and -0.56783 values between -0.56 and -0.052542 values between -0.05 and 0.46340 values between 0.46 and 0.9719 values between 0.97 and 1.4810 values between 1.48 and 1.998 values between 1.99 and 2.504 values between 2.50 and 3.017 values between 3.01 and 3.521 values between 3.52 and 4.031 values between 4.54 and 5.05offnegativepositiveoften

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. تف
  2. Cran
  3. specialchars
  4. 车型
  5. Vit
  6. */,⏎
  7. Carolyn
  8. ادة
  9. Acer
  10. ホーム

Suppresses

  1. izzato
  2. 印象
  3. 的印象
  4. 记
  5. ksi
  6. 嗓子
  7. ana
  8. unate
  9. unu
  10. rias