Neuron Explorer

Qwen3-0.6B

linking words like 'further' and 'worse'

A neuron in layer 25 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 responds to small connecting words that introduce a follow-up point, such as 'worse', 'better' and 'further'. Turning it up made the model keep producing the words 'still' and 'yet' over and over.

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

    literally hundreds of formats for graphical images. Although there are a few dozen formats publicly standardized, most are bitmapped formats -- especially hard to collaborate on -- and, worse, many incorporate data compression. Most of the rest require proprietary tools or operating systems

  2. #2 peak

    't complain (108).⏎7: Correct Usage 8 | An unlikely row has erupted in France over suggestions that the semicolon's days are numbered; worse, the growing influence of English is apparently to blame.⏎8: Correct Usage

  3. #3 peak

    Or better said: They should eat meat only.⏎Ferrets have a particular diet consisting

  4. #4 peak

    , it’s safe to say that there is no evidence that cell phones increase your chance of getting a brain tumor. Further studies will surely be conducted in the coming years but at this point, you’re

  5. #5 peak

    ould, who were collaborating with Frederick Law Olmsted in structures for Central Park, was soon eclipsed by the South range of the museum, by J. Cleaveland Cady, a robust exercise in rusticated brownstone neo-Roman

  6. #6 peak

    Cincinnati occurred in 1788 at the mouth of the Little Miami. Early entrepreneurs Israel Ludlow, Matthias Denman, and Robert Patterson purchased from John Cleves Symmes 800 acres of land and created the second settlement opposite

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: what is the more recent than the more recent? What is the more recent than the more recent than the more recent than... going on and on? What is the more recent than the more recent than the more recent than the more recent than

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

    -edge techniques. Therefore, European industries involved in the development of sustainable seaweed aquaculture need to be supported.⏎Authors: Michèle Barbier, Bénédicte Charrier, Rita Araujo,

  2. #2 peak

    development of emotional intelligence in young children. As Robert Emmons, Ph.D. the world’s leading scientific expert on gratitude. a professor of psychology and founding editor-in-chief of The Journal of Positive Psychology, has discovered when people regularly engage in gratitude

  3. #3 peak

    thus ensuring tranquillity during much of his reign. In his final years, however, he faced internal disturbances, and he was ultimately killed in an attempted coup d’état.⏎Son of Setnakht (reigned 1190

  4. #4 peak

    He changed his approach immediately.⏎He was not a player in the October Revolution.⏎The Bolshevik revolution⏎Three interpretations of the October Revolution: popular uprising, coup d'etat

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. L16:1733 weakly excites
  2. L18:650 weakly suppresses the word 'more' before a comparison
  3. L17:849 weakly excites
  4. L18:1454 weakly suppresses
  5. L8:14 weakly excites
  6. L13:2501 weakly suppresses
  7. L18:3065 weakly excites
  8. L17:2344 weakly suppresses
  9. L15:1259 weakly suppresses
  10. L13:1416 weakly suppresses
  11. L24:8 weakly suppresses
  12. L8:7 weakly excites
  13. L8:3 weakly excites
  14. L8:12 weakly excites
  15. L17:1900 weakly excites
  16. L4:92 weakly excites
  17. L16:248 weakly excites
  18. L22:535 weakly excites
  19. L1:1663 weakly excites
  20. L1:2133 weakly suppresses
  21. L4:1614 weakly suppresses
  22. L12:634 weakly excites
  23. L23:2405 weakly excites
  24. L8:44 weakly excites

Feeds later layers

  1. L27:1280 strongly excites
  2. L27:2009 weakly excites
  3. L26:2595 weakly excites
  4. L27:2489 weakly suppresses
  5. L27:644 weakly excites
  6. L27:1647 weakly suppresses
  7. L26:2690 weakly excites
  8. L26:814 weakly excites
  9. L27:1878 weakly suppresses
  10. L26:2996 weakly excites
  11. L27:572 weakly suppresses
  12. L27:2570 weakly suppresses 'con' starting a word piece
  13. L26:1262 weakly suppresses
  14. L26:1994 weakly suppresses
  15. L27:820 weakly suppresses
  16. L27:461 weakly suppresses
  17. L27:1254 weakly excites
  18. L27:1123 weakly excites
  19. L27:849 weakly suppresses
  20. L27:680 weakly suppresses
  21. L27:1132 weakly suppresses
  22. L27:3045 weakly suppresses capital W starting a word
  23. L26:1909 weakly suppresses
  24. L27:1341 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.

2 values between -9.73 and -8.871 values between -8.87 and -8.011 values between -8.01 and -7.153 values between -7.15 and -6.305 values between -6.30 and -5.443 values between -5.44 and -4.587 values between -4.58 and -3.724 values between -3.72 and -2.867 values between -2.86 and -2.0064 values between -2.00 and -1.151232 values between -1.15 and -0.292373 values between -0.29 and 0.57125 values between 0.57 and 1.4326 values between 1.43 and 2.297 values between 2.29 and 3.1411 values between 3.14 and 4.009 values between 4.00 and 4.864 values between 4.86 and 5.725 values between 5.72 and 6.584 values between 6.58 and 7.433 values between 7.43 and 8.295 values between 9.15 and 10.012 values between 10.01 and 10.872 values between 11.72 and 12.582 values between 12.58 and 13.441 values between 23.74 and 24.59offnegativepositiveoften

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. still
  2. still
  3. Still
  4. STILL
  5. Still
  6. 仍然
  7. yet
  8. 仍
  9. 依然
  10. 还是

Suppresses

  1. ave
  2. 第二节
  3. こと
  4. /uploads
  5. เฉพาะ
  6. inheritdoc
  7. enders
  8. andom
  9. (loop
  10. 楸