Neuron Explorer

Qwen3-0.6B

'out' in 'turns out'

A neuron in layer 4 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 lights up on the word 'out' in phrases like 'it turns out', where a surprising fact is about to be revealed. It has a strong opposite side on the word 'touch' in 'get in touch', which pushes it the other way. Turning it up did not give a clear theme in the model's replies.

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

    structures that are more efficient to run, cutting ongoing costs and minimizing the carbon footprint.⏎What does this have to do with BrightBuilt Barn?⏎Well, it turns out that the Barn was also designed with (simpler and much less expensive)

  2. #2 peak

    pain relief from morphine?⏎Zaborina and Alverdy decided to catalog the surviving gut microbes, find out who they are and—more importantly it turns out—how they are behaving in these patients at risk for sepsis.⏎They

  3. #3 peak

    in domestic rubbish can only be gasified at low temperatures, but this in turn creates tars and waxes that stifle the process.⏎However, it turns out that these tars and waxes are no match for an 800

  4. #4 peak

    -counter medication someone has taken. The toilet can also identify different users by scanning the one part of the anatomy that is always presented to the bowl. (It turns out that this particular part of the anatomy is as individual as a fingerprint. Who knew

  5. #5 peak

    a period that has been dubbed the “magic moment.” The new study is the first to test that assumption, said author Christina Gibson-Davis. “It turns out the ‘magic moment’ lasts longer than conventional wisdom has held,” said Gibson-D

  6. #6 peak

    . But I was still floored. How could a local government agency move so quickly?⏎Turns out, it didn’t. In 2007, Google engineers asked public

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 user is a member of the community, and I have to provide for the community with which I have to provide with the community, and I have to get in touch with someone. The user has to be given with the community to be

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

    Python’s chanting Vikings drowned aside most other conversation into the the latest eatery, the advertisements began drowning away normal communication online. Thus, anybody already been getting in touch with unwanted . Etymology, such as negation, try rarely used by yourself

  2. #2 peak

    students not to talk to strangers or accept rides from someone they don’t know.⏎- Make sure the student knows their phone number, address, how to get in touch with their parents at work, how to get in touch with another trusted adult and

  3. #3 peak

    sensation to a chart-topping music genius lauded for his complex rhythms and socially-conscious lyrics about racism, black liberation, love and unity. He had kept in touch

  4. #4 peak

    Now the television is in competition with other devices -- computers, video games and smartphones. All of these screens offer entertainment, information, and a way to keep in touch with and stay connected to others.⏎With the arrival of the coronavirus pandemic, screens

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. L3:50 excites
  2. L3:1806 weakly excites
  3. L1:2278 weakly excites
  4. L0:1958 weakly excites
  5. L3:2369 weakly suppresses
  6. L3:1926 weakly suppresses
  7. L3:464 weakly excites
  8. L3:2548 weakly suppresses
  9. L2:1367 weakly excites
  10. L3:1287 weakly suppresses
  11. L2:237 weakly excites
  12. L3:2810 weakly suppresses
  13. L2:2113 weakly suppresses
  14. L3:342 weakly excites
  15. L1:1255 weakly suppresses
  16. L1:1917 weakly suppresses
  17. L1:994 weakly suppresses
  18. L2:784 weakly excites
  19. L3:2044 weakly excites
  20. L2:2075 weakly suppresses
  21. L3:2635 weakly excites
  22. L1:1229 weakly excites
  23. L1:2179 weakly excites
  24. L2:92 weakly excites

Feeds later layers

  1. L7:2802 weakly suppresses
  2. L10:1418 weakly suppresses
  3. L10:2474 weakly suppresses
  4. L17:305 weakly excites
  5. L5:2534 weakly suppresses
  6. L6:670 weakly excites
  7. L20:1335 weakly suppresses
  8. L7:2462 weakly excites
  9. L9:231 weakly excites muscle and body movement words
  10. L17:2135 weakly suppresses 'the' after 'of' and similar words
  11. L11:445 weakly excites
  12. L5:2766 weakly excites
  13. L7:2039 weakly suppresses
  14. L11:2186 weakly excites
  15. L5:2340 weakly excites
  16. L13:35 weakly suppresses
  17. L22:2730 weakly suppresses
  18. L12:1050 weakly suppresses
  19. L8:2327 weakly excites
  20. L8:1939 weakly excites
  21. L17:1466 weakly suppresses
  22. L7:2165 weakly suppresses
  23. L13:182 weakly suppresses
  24. L9:711 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 -3.85 and -3.721 values between -2.92 and -2.781 values between -2.78 and -2.652 values between -1.18 and -1.052 values between -1.05 and -0.912 values between -0.91 and -0.782 values between -0.64 and -0.5112 values between -0.51 and -0.3872 values between -0.38 and -0.24412 values between -0.24 and -0.111858 values between -0.11 and 0.021190 values between 0.02 and 0.16297 values between 0.16 and 0.2945 values between 0.29 and 0.434 values between 0.43 and 0.563 values between 0.69 and 0.831 values between 0.96 and 1.092 values between 1.09 and 1.231 values between 1.36 and 1.50offnegativepositiveoften

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. side
  2. avor
  3. DEAL
  4. ilar
  5. lado
  6. nom
  7. OUS
  8. SIDE
  9. 梧
  10. 凳

Suppresses

  1. 致电
  2. ��
  3. mails
  4. 第一书记
  5. SystemService
  6. 热线
  7. hotline
  8. بتاريخ
  9. 骚扰
  10. RelayCommand