Neuron Explorer

Qwen3-0.6B

the word 'blood' and its compounds

A neuron in layer 23 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 'blood' when it appears in phrases like 'blood clot', 'bloodline' or 'blood cells'. When it was turned up, the model's replies drifted into odd, unrelated text and did not clearly show the blood theme.

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

    from coming together. Whenever there is an injury in your body, platelets are sent to the site of the injury, where they clump together to form a blood clot. This stops the bleeding in your body. If you have a cut or

  2. #2 peak

    similar-looking poisonous plants, among them poison hemlock and fool's parsley ... Cow parsley can be confused with giant cow parsley/giant hogweed (Heracleum mantegazzianum), the sap of which can cause severe burns after coming in

  3. #3 peak

    used to thin the blood of a patient allowing the blood to flow more easily through the veins. This means that the blood will be less likely to form a dangerous blood clot. Warfarin is a commonly prescribed anticoagulant medicine. Min

  4. #4 peak

    In embryonic development, stem cells differentiate into organs by means of the actions of morphogens. And even cancer cells can use morphogens to grow, induce a blood supply, and metastasize.⏎Although the concept of cell organization in response to

  5. #5 peak

    in the immune system, thus increasing the metabolic rate, the release of some hormones and growth factors. A short exercise session raises the counts of neutrophils, blood cells that play an important role in fighting

  6. #6 peak

    Alaska and Russia in massive numbers – these land haul outs result in trampling of their young,” L<|endoftext|>What are ticks and where do they live?⏎Ticks are blood-feeding parasites that are often found in tall grass where they will wait to attach

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: At the start of each operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator operator

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

    from internal and external influences.⏎In the 25th April 1974, some militaries started a coup against the government of Marcello Caetano. They protested against the lack of democracy, the censorship and the lack of

  2. #2 peak

    overall memory which means that multiple language speakers The Positive Negative Effects of Learning a Learning a foreign language is a brave decision on the part of the learner<|endoftext|>n.交響樂 [an elaborate instrumental composition in three or more movements, similar in form

  3. #3 peak

    the American Armed Forces invaded the country in order to keep the régime in place. The leaders of the revolution were either killed or arrested to be later executed. Caetano returned to power. The people were again in the darkness.⏎Creation of<|endoftext|>

  4. #4 peak

    instead of the two overloaded versions provided in C++.⏎/* memchr example */⏎int main ()⏎char * pch;⏎char str = "Example string";⏎pch = (char

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. L19:1978 excites
  2. L22:530 suppresses
  3. L3:2548 weakly suppresses
  4. L16:1295 weakly suppresses
  5. L20:2428 weakly excites
  6. L20:1120 weakly excites
  7. L8:328 weakly excites
  8. L20:1180 weakly excites
  9. L21:1367 weakly suppresses
  10. L13:2616 weakly suppresses
  11. L15:59 weakly suppresses
  12. L10:1801 weakly suppresses
  13. L15:2540 weakly suppresses
  14. L2:936 weakly suppresses
  15. L3:1311 weakly suppresses
  16. L17:292 weakly excites
  17. L17:400 weakly excites
  18. L12:1038 weakly suppresses
  19. L5:169 weakly excites
  20. L18:569 weakly excites
  21. L17:1406 weakly suppresses
  22. L2:551 weakly suppresses
  23. L4:1233 weakly excites
  24. L9:512 weakly excites

Feeds later layers

  1. L27:1319 suppresses
  2. L24:1834 weakly excites
  3. L26:1676 weakly suppresses
  4. L26:1703 weakly suppresses
  5. L24:1501 weakly excites
  6. L27:2779 weakly excites
  7. L27:434 weakly excites
  8. L27:599 weakly suppresses
  9. L24:1762 weakly excites
  10. L25:739 weakly suppresses
  11. L26:2585 weakly suppresses
  12. L25:18 weakly suppresses
  13. L26:2662 weakly excites
  14. L24:696 weakly excites
  15. L26:2842 weakly excites
  16. L26:1679 weakly excites
  17. L25:2284 weakly excites
  18. L24:1447 weakly excites
  19. L24:90 weakly excites
  20. L25:111 weakly excites comma before a 'because' or 'as' clause
  21. L27:2491 weakly suppresses
  22. L25:2843 weakly excites
  23. L26:2877 weakly suppresses
  24. L25:771 weakly suppresses

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 -4.71 and -4.151 values between -4.15 and -3.603 values between -2.49 and -1.9312 values between -1.93 and -1.3847 values between -1.38 and -0.82446 values between -0.82 and -0.273247 values between -0.27 and 0.29116 values between 0.29 and 0.8513 values between 0.85 and 1.404 values between 1.40 and 1.965 values between 1.96 and 2.513 values between 2.51 and 3.071 values between 3.62 and 4.182 values between 4.73 and 5.291 values between 5.84 and 6.403 values between 8.06 and 8.621 values between 10.28 and 10.841 values between 11.95 and 12.501 values between 16.95 and 17.50offnegativepositiveoften

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. tring
  2. 着他
  3. ิง
  4. 统
  5. raft
  6. inté
  7. füh
  8. he
  9. Leben
  10. 锾

Suppresses

  1. operator
  2. →operator
  3. operator
  4. operators
  5. (operator
  6. .operator
  7. _operator
  8. Operators
  9. Operator
  10. -operator