Neuron Explorer

Qwen3-0.6B

unclear: word before 'first', 'last' or 'further'

A neuron in layer 17 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on words that often lead into a sequence or key point, like 'the first', 'the last' or 'further'. The pattern is loose and hard to name. Turning it up made the model produce repeated words and half-finished phrases.

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

    animal protein causes a lining of fat to form in their stomach walls. This lining of fat affects their absorption rate which can end up becoming dangerous to them. Aside from containing animal protein, dog food also contains meat.

  2. #2 peak

    in your professional development<|endoftext|>Mackinder’s Heartland Theory made SIMPLE⏎Do you want to learn about Mackinder’s Heartland Theory? Then you have come to the right

  3. #3 peak

    PRAYER AND PRAISE (5 minutes)⏎Further info? Click here⏎MEMORY WORK (12 minutes)⏎"You will seek me and find me when you seek me with all your heart" Jeremiah 29:13⏎Write the

  4. #4 peak

    the “unsinkable” Titanic set sail from Southhampton, England on April 10, 1912.⏎This is one of the last photographs of Titanic taken from the shore as the ship departed towards New York City.⏎

  5. #5 peak

    the body. The solution contains chemicals such as⏎The mixture of these solutions allows the body to last longer than it could have stayed in normal conditions. They first start by drawing blood from the body using the veins. After this, they inject the

  6. #6 peak

    because the typical symptoms of an acoustic neuroma are non-specific and can often be put down to other causes. Symptoms include:⏎- Hearing loss (this is the first symptom of an acoustic neuroma in around 90% of cases)⏎-

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 have a 30000000000000000000000000000000000000000000

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

    Zeus, lord of the gods, who rapes his sister Demeter, and the product of that rape is Persephone. They never married. Indeed, Zeus would

  2. #2 peak

    top scholars discuss the historical Israelites in Egypt and archaeological evidence for and against the historicity of the Exodus.⏎BAR coauthor Lily Singer-Avitz suggests that several finds discovered in the later strata,

  3. #3 peak

    . At least, that’s what we thought we would be doing. Continue reading →⏎Black pepper and its flavoring companion, salt, can be found in kitchens all around the world. But did you

  4. #4 peak

    seems appropriate to have some humility around this energy transition,” said co-author David Rapson, an associate professor at the UC Davis Economics Department. “Appro

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. L13:1971 suppresses
  2. L11:1383 excites
  3. L12:2475 suppresses
  4. L8:968 excites
  5. L7:2458 excites
  6. L16:961 suppresses
  7. L11:2850 suppresses
  8. L7:2553 suppresses
  9. L12:1705 suppresses
  10. L4:1724 excites
  11. L14:1373 excites
  12. L13:2562 excites
  13. L16:1471 excites
  14. L10:2453 suppresses 'of' after publisher or division nouns
  15. L14:623 excites
  16. L5:2500 excites
  17. L14:2671 suppresses
  18. L9:755 weakly suppresses
  19. L9:906 weakly excites
  20. L14:526 weakly suppresses
  21. L4:929 weakly suppresses
  22. L9:731 weakly excites
  23. L4:869 weakly excites
  24. L9:1705 weakly suppresses

Feeds later layers

  1. L18:1812 suppresses
  2. L18:976 suppresses
  3. L18:1672 weakly excites
  4. L18:919 weakly suppresses
  5. L19:1235 weakly excites
  6. L18:2621 weakly suppresses
  7. L18:759 weakly excites
  8. L18:1397 weakly excites
  9. L20:1592 weakly suppresses
  10. L22:781 weakly suppresses
  11. L23:553 weakly suppresses
  12. L21:1974 weakly excites
  13. L23:2021 weakly excites
  14. L18:1624 weakly suppresses
  15. L24:594 weakly suppresses
  16. L21:1420 weakly excites
  17. L23:1785 weakly suppresses
  18. L19:1786 weakly excites
  19. L20:2830 weakly excites
  20. L18:456 weakly excites
  21. L24:1231 weakly excites
  22. L24:2098 weakly excites
  23. L19:462 weakly suppresses
  24. L24:1411 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 -19.09 and -18.491 values between -17.28 and -16.681 values between -11.85 and -11.241 values between -11.24 and -10.642 values between -10.04 and -9.432 values between -9.43 and -8.831 values between -8.83 and -8.233 values between -8.23 and -7.624 values between -7.62 and -7.022 values between -7.02 and -6.412 values between -6.41 and -5.812 values between -5.81 and -5.214 values between -5.21 and -4.6010 values between -4.60 and -4.0011 values between -4.00 and -3.4015 values between -3.40 and -2.7918 values between -2.79 and -2.1937 values between -2.19 and -1.5886 values between -1.58 and -0.98260 values between -0.98 and -0.382400 values between -0.38 and 0.23846 values between 0.23 and 0.83131 values between 0.83 and 1.4442 values between 1.44 and 2.0417 values between 2.04 and 2.645 values between 2.64 and 3.251 values between 3.25 and 3.852 values between 3.85 and 4.461 values between 4.46 and 5.06offnegativepositiveoften

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. au
  2. adget
  3. anson
  4. фор
  5. rior
  6. arbon
  7. 俦
  8. ốt
  9. ybrid
  10. esterday

Suppresses

  1. ije
  2. 虱
  3. elen
  4. ships
  5. S
  6. anj
  7. primaries
  8. 舍
  9. anine
  10. multipart