Neuron Explorer

Qwen3-0.6B

'h' fragments and 'error' words

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 lights up on a few word pieces, mainly a lone 'H' or '-h' inside words like long-haul and hurricane, and on the word 'error'. The negative side fires on 'H' at the start of gene and protein names. When turned up, the writing drifted toward words like 'exception', though the effect was loose.

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

    Goodreads helps you keep track of books you want to read. Want to Read saving…. Want to Read Currently Reading Read. Other editions. Enlarge cover. Error rating book.⏎Browse more videos⏎Human Anatomy and Physiology are two most-related subjects

  2. #2 peak

    think, why ye are just the sen!⏎– Singin’ blow, ye winds, in the mornin’,⏎– Blow, ye winds, high-ho!⏎– See all clear yer runnun’ gear,⏎– An’

  3. #3 peak

    advantage because it would provide a strong base of resources and manpower that could<|endoftext|>LISE SCHIFFER, LCSW⏎Your Cart is Empty⏎There was an error with PayPalClick here to try again⏎Thank you for

  4. #4 peak

    the benefits of no contrails.⏎However, global air traffic is growing by around 3.5% each year, and many of these new flights are long-haul, high-altitude trips, which are the kind that form the most contr

  5. #5 peak

    had experienced severe disease requiring intensive care treatment.⏎In this way, Gao and his colleagues demonstrated that the tool could reveal signs of pulmonary fibrosis in COVID long-haulers, thus helping to account for shortness of breath, coughing and

  6. #6 peak

    a typhoon if it happened in the northwestern Pacific.⏎WHAT’S THE DIFFERENCE BETWEEN A HURRICANE AND A TROPICAL STORM?⏎Wind speed. When tropical

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: not all of the physical contents of a notepartepel would go into physical contents that would not come into physical contents yet, not all of the physical contents of a notepartepel would go into physical contents that would not

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

    genes have been identified that increase the risk of prostate cancer, including RNASEL, BRCA 1, and BRCA 2, DNA mismatch genes, HPC1, and HoxB13

  2. #2 peak

    cells of mesenchymal origin. Upon HGF stimulation, MET induces several biological responses that collectively give rise to a program known

  3. #3 peak

    cause devastating side effects to children, and it's not clear how beneficial the treatments are for this disease.⏎Researchers are also exploring a potential new therapy that could target H3K27me3 and reverse the lowered levels. This work is still

  4. #4 peak

    by testing them for anti-HIV activity. In test tube experiments called “assays,” new compounds are added

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. L22:1086 excites
  2. L22:1061 suppresses
  3. L19:2916 suppresses
  4. L20:1514 weakly suppresses
  5. L1:19 weakly suppresses
  6. L23:1073 weakly suppresses
  7. L22:693 weakly suppresses
  8. L24:105 weakly suppresses
  9. L20:359 weakly excites
  10. L22:1950 weakly excites
  11. L9:1932 weakly excites
  12. L9:2375 weakly suppresses
  13. L1:87 weakly suppresses
  14. L21:2904 weakly excites
  15. L18:344 weakly suppresses
  16. L21:3013 weakly suppresses
  17. L16:1235 weakly suppresses
  18. L18:173 weakly excites
  19. L11:1325 weakly suppresses
  20. L2:298 weakly suppresses
  21. L21:421 weakly excites
  22. L0:2569 weakly suppresses
  23. L9:1267 weakly excites
  24. L0:1686 weakly suppresses

Feeds later layers

  1. L27:1901 weakly excites
  2. L26:1473 weakly excites
  3. L27:202 weakly excites
  4. L26:2604 weakly suppresses
  5. L27:1438 weakly suppresses
  6. L27:714 weakly suppresses
  7. L26:250 weakly excites
  8. L26:1700 weakly excites
  9. L26:1782 weakly suppresses
  10. L27:1153 weakly excites
  11. L26:2646 weakly suppresses
  12. L26:1285 weakly excites
  13. L27:466 weakly suppresses
  14. L27:1505 weakly excites
  15. L27:1591 weakly suppresses
  16. L26:2119 weakly suppresses
  17. L27:1040 weakly suppresses
  18. L26:2901 weakly suppresses
  19. L27:1355 weakly suppresses
  20. L27:1631 weakly excites
  21. L27:2834 weakly excites
  22. L27:1859 weakly excites
  23. L27:923 weakly suppresses
  24. L26:1452 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 -11.75 and -11.161 values between -11.16 and -10.561 values between -7.00 and -6.412 values between -5.82 and -5.222 values between -4.63 and -4.042 values between -4.04 and -3.441 values between -3.44 and -2.854 values between -2.85 and -2.265 values between -2.26 and -1.669 values between -1.66 and -1.07120 values between -1.07 and -0.482955 values between -0.48 and 0.12735 values between 0.12 and 0.7150 values between 0.71 and 1.304 values between 1.30 and 1.901 values between 1.90 and 2.494 values between 2.49 and 3.083 values between 3.08 and 3.681 values between 3.68 and 4.272 values between 5.46 and 6.052 values between 6.05 and 6.641 values between 7.24 and 7.831 values between 9.02 and 9.611 values between 11.39 and 11.98offnegativepositiveoften

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. thrown
  2. catching
  3. exception
  4. Catch
  5. .exception
  6. thrown
  7. edException
  8. �行
  9. ERN
  10. ainers

Suppresses

  1. mor
  2. cimal
  3. ycin
  4. ide
  5. 垩
  6. 关节
  7. itably
  8. 价值
  9. мор
  10. ises