Neuron Explorer

Qwen3-0.6B

'therefore' or 'thus' after a subject

A neuron in layer 10 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 fires on words like 'therefore' or 'thus' that come right after the subject of a sentence in formal, science-style writing, where a conclusion is about to follow. Turning it up made the model's text drift toward unrelated question-and-answer style sentences.

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

    advantage of this unique research platform to measure greenhouse gases and aerosols, such as particulate matter,⏎as well as to verify Dutch and European emissions. It is therefore making a key contribution to two European research infrastructures: ICOS

  2. #2 peak

    CHR) in minor head injury is an evidence-based aid in decision making as regards to use of CT-scans to detect head injury requiring neuro-intervention. It therefore avoids wastage of resources. The objective was to compare the number of CT-sc

  3. #3 peak

    the hepatic vein.⏎It should be noted that during the process of clearing the body of toxins, the liver is damaged by exposure to harmful substances. The liver therefore possesses the amazing ability to self-repair and regenerate of damaged tissues.⏎So as

  4. #4 peak

    as "Lunda-designs". The Tchokwe are descendants of the Lunda people, and live predominantly in a region today called Lunda, hence the⏎name Lunda-design to honour my source of inspiration.⏎The picture below presents

  5. #5 peak

    canal treatment) is a dental technique that provides dead or inflamed teeth. The bacteria are removed with special instruments and the function of the tooth is restored. He therefore does not have to be drawn. Successful root canal treatment can preserve the tooth for

  6. #6 peak

    considerable economic importance. They are built in their hundreds of thousands every year, and have a history extending back to the 1920s. It is therefore surprising there have been so few books dealing in any depth with solid-state

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 know that, for example, in a certain kind of mathematical structure, that is, that is, that is the structure of, that is, that is the structure of a group, that is, that is the structure of a group

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

    fracture and it becomes much more straightforward for health care professionals to follow an appropriate standard of care in order to best treat the injury. To provide you with more information so that you may better understand whiplash injuries and how they should be treated,

  2. #2 peak

    of capturing the beasts is recorded in several Roman mosaics.⏎The cookie settings on this website are set to "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing your cookie settings or you

  3. #3 peak

    Collagen, five primary ones are the most important here.⏎Type I: This type accounts for 90% of the collagen in your body. These give form and structure to the bones with calcium, skin, arteries, tendons, scar

  4. #4 peak

    , the U.S. Department of Agriculture Food Safety and Inspection Service warns against using these as a fruit and vegetable wash for your fresh

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. L7:1112 strongly excites
  2. L6:2530 strongly suppresses
  3. L7:2662 excites
  4. L6:2753 suppresses
  5. L4:1529 excites
  6. L6:2820 suppresses
  7. L7:1253 suppresses
  8. L9:212 suppresses
  9. L7:2470 suppresses
  10. L7:625 excites
  11. L9:298 suppresses
  12. L5:1035 excites
  13. L7:104 excites
  14. L7:627 excites
  15. L8:1885 weakly suppresses
  16. L3:2565 weakly excites
  17. L4:386 weakly excites
  18. L6:2050 weakly suppresses
  19. L6:1575 weakly excites
  20. L7:1234 weakly suppresses
  21. L6:1485 weakly excites
  22. L7:912 weakly suppresses
  23. L6:1802 weakly suppresses
  24. L8:133 weakly suppresses

Feeds later layers

  1. L15:1116 excites
  2. L14:665 suppresses
  3. L15:2554 excites
  4. L13:1727 excites
  5. L12:2704 excites
  6. L15:3012 excites
  7. L12:144 suppresses
  8. L13:3023 excites
  9. L12:2151 excites
  10. L15:2624 excites
  11. L15:516 suppresses
  12. L16:1241 excites
  13. L12:2436 excites
  14. L16:2425 excites
  15. L19:3020 excites
  16. L13:181 suppresses
  17. L17:3032 excites
  18. L16:2177 excites
  19. L15:1854 excites
  20. L14:628 excites end of document boundary token
  21. L13:804 excites
  22. L19:1478 suppresses
  23. L12:2968 weakly excites
  24. L17:1448 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.

87 values between -0.90 and -0.65705 values between -0.65 and -0.401865 values between -0.40 and -0.15997 values between -0.15 and 0.1090 values between 0.10 and 0.3551 values between 0.35 and 0.6024 values between 0.60 and 0.8517 values between 0.85 and 1.1016 values between 1.10 and 1.368 values between 1.36 and 1.618 values between 1.61 and 1.869 values between 1.86 and 2.115 values between 2.11 and 2.361 values between 2.36 and 2.612 values between 2.61 and 2.863 values between 2.86 and 3.113 values between 3.11 and 3.365 values between 3.36 and 3.611 values between 3.61 and 3.862 values between 3.86 and 4.121 values between 4.12 and 4.372 values between 4.37 and 4.621 values between 5.37 and 5.621 values between 5.87 and 6.121 values between 6.88 and 7.131 values between 7.38 and 7.631 values between 7.88 and 8.131 values between 8.88 and 9.13offnegativepositiveoften

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. ffield
  2. :title
  3. entially
  4. Lilly
  5. .addField
  6. ettle
  7. getField
  8. :url
  9. xDD
  10. 走上

Suppresses

  1. 又是
  2. 又能
  3. _again
  4. 貫
  5. |
  6. 件
  7. itori
  8. uno
  9. enade
  10. ograms