Neuron Explorer

Qwen3-0.6B

'the' after a preposition

A neuron in layer 19 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 fires on 'the' that comes right after a preposition, as in 'with the virus' or 'production of the 60 watt bulb'. It goes quiet or negative on the preposition itself, like 'of' or 'by' just before a name. The steered text it produced was garbled, so it does not 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

    wondered just how safe the cloud is? In this article, we explore how secure the cloud is and how cloud security can be improved.⏎What is the Cloud?⏎The cloud, in the context of business operations, refers to a network of remote servers

  2. #2 peak

    you on high alert. But don’t let worries about mosquitoes keep you and your family from enjoying the outdoors. The chances of anyone in your family becoming sick with the West Nile virus are extremely low, no matter what part of the country you live

  3. #3 peak

    receive. It increases vitamin D and now we can see it improves nitric oxide levels.⏎Why does this matter to the coronavirus? The people who are hospitalized with the virus have a host of other health problems, chief amongst them is high blood pressure

  4. #4 peak

    eral ring (shown here in the day- and night-active dinosaur, Saurolophus osborni). Diurnal, or day-active, animals with the ring have a small opening, whereas nocturnal animals have a

  5. #5 peak

    SUPPORT RECIPE from Lilsipper | Digital Download 7 Day Vegan Meal Plan | Digital download⏎Awesome Benefits Of The Paleo Diet⏎The idea behind the paleo diet is that we haven’t changed much since the time of the earliest

  6. #6 peak

    7 -the elimination of the incandenscent light bulb as we know it. By the year 2014 we will not see the production of the 60 watt light bulb which we have become so familiar with. It will

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 following is a possible example of the same thing, but with different words. What should be the other example of the same thing, with different words, and with the same meaning, and with other words, and other words, and with

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

    Formula One (also Formula 1 or F1 and officially the FIA Formula One World Championship) is the highest class of single-seat auto racing that is sanctioned by the Fédération Internationale de l'Automobile (FIA). The

  2. #2 peak

    protect US government communications and information systems from foreign intelligence and security threats.⏎Does the CIA fall under the Department of Defense?⏎The majority of US intelligence agencies are part of<|endoftext|>The ability to solve word problems ranks high on any math teacher’s list of

  3. #3 peak

    for the interested and curious that will hopefully inspire further reading.<|endoftext|>Today is Flag Day.⏎"June 14 is Flag Day, which commemorates the adoption of the flag

  4. #4 peak

    the group changed its name to the STEMG.⏎The STEMG reports to the ASD Product Support Specification Group (PSSG) which in turn reports to the ASD Services Commission (SVC).<|endoftext|>Biology exquisitely creates

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. L18:1264 suppresses
  2. L18:861 excites
  3. L18:1790 suppresses
  4. L18:1084 excites
  5. L13:523 excites
  6. L11:226 excites
  7. L18:1990 weakly excites
  8. L17:1557 weakly suppresses
  9. L18:1421 weakly suppresses
  10. L1:732 weakly suppresses
  11. L18:172 weakly suppresses
  12. L6:234 weakly excites
  13. L16:216 weakly excites
  14. L1:120 weakly excites
  15. L17:372 weakly excites
  16. L17:1194 weakly excites
  17. L2:1372 weakly suppresses
  18. L17:908 weakly suppresses
  19. L18:1498 weakly suppresses
  20. L15:1931 weakly excites
  21. L11:580 weakly excites
  22. L18:1041 weakly excites
  23. L15:1492 weakly suppresses
  24. L16:1083 weakly excites

Feeds later layers

  1. L20:955 strongly excites
  2. L20:439 excites
  3. L21:653 excites
  4. L20:769 excites
  5. L23:24 suppresses
  6. L21:99 suppresses
  7. L20:181 suppresses
  8. L20:2072 excites
  9. L20:1195 excites
  10. L20:1112 excites
  11. L21:2754 excites
  12. L20:1677 excites
  13. L20:668 excites
  14. L22:241 excites
  15. L20:385 excites
  16. L20:176 excites
  17. L21:713 weakly suppresses
  18. L20:88 weakly excites
  19. L20:956 weakly excites
  20. L20:288 weakly excites
  21. L20:482 weakly excites
  22. L20:1131 weakly excites
  23. L21:775 weakly excites the article 'a' before a noun
  24. L21:404 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 -10.71 and -10.232 values between -8.77 and -8.281 values between -7.80 and -7.313 values between -7.31 and -6.836 values between -6.83 and -6.341 values between -6.34 and -5.865 values between -5.86 and -5.372 values between -5.37 and -4.893 values between -4.89 and -4.4017 values between -4.40 and -3.9216 values between -3.92 and -3.4315 values between -3.43 and -2.9524 values between -2.95 and -2.4639 values between -2.46 and -1.9856 values between -1.98 and -1.4991 values between -1.49 and -1.00193 values between -1.00 and -0.521119 values between -0.52 and -0.031895 values between -0.03 and 0.45171 values between 0.45 and 0.9465 values between 0.94 and 1.4237 values between 1.42 and 1.9134 values between 1.91 and 2.3931 values between 2.39 and 2.8822 values between 2.88 and 3.3616 values between 3.36 and 3.8514 values between 3.85 and 4.3411 values between 4.34 and 4.827 values between 4.82 and 5.316 values between 5.31 and 5.792 values between 5.79 and 6.282 values between 6.28 and 6.761 values between 8.22 and 8.70offnegativepositiveoften

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. \Collection
  2. ensively
  3. estar
  4. ";⏎
  5. GLOBALS
  6. ",",
  7. "go
  8. Hague
  9. ETERS
  10. egie

Suppresses

  1. unks
  2. ←
  3. dee
  4. medic
  5. 疽
  6. 核定
  7. ude
  8. ثر
  9. dd
  10. Raq