Neuron Explorer

Qwen3-0.6B

subject pronouns inside subordinate clauses

A neuron in layer 24 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

The neuron fires on the subject pronoun that starts a clause after words like 'because', 'while', 'knew' or 'thought', such as 'because he' or 'I didn't know I'. Turning it up did not give a clear theme in the output.

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

    century evaluations about jesuitic missions among the Guaranis - Relação abbreviada da republica, que os religiosos jesuitas estabelecerão no Paraguai, by Marquis of Pombal, Cand

  2. #2 peak

    gene flow patterns among Tasmanian devil populations, Biological Conservation, 165 pp. 69-78. ISSN 0006-320

  3. #3 peak

    more often than non-carriers, they failed to remember to write down the desired word when they were supposed to – in other words, they forgot to do what they meant to do, when they meant to do it.⏎Because the Alzheimer’s genotype

  4. #4 peak

    a cross that is used as a symbol for Christianity.⏎definition 3: a difficulty or burden.⏎- Being teased by my brothers was my cross to bear while I was growing up.⏎synonyms: adversity, affliction, burden, distress,

  5. #5 peak

    groups are ortho- and para- directing, and all deactivating groups other than halogen are meta directing. The halogens<|endoftext|>I didn’t know I had access to the entire back catalog of The New York Times through my library card

  6. #6 peak

    dedicated student. He even slept with wine corks between his fingers because he was told it might help him achieve a wider span on the piano. By the

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: some people are in some way, in some way, in some way, in some way, in some way, in some way, in some way, in some way, in some way, in some way, in some way, in

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

    stores had responded to the protesters by ignoring them or closing down the lunch counters entirely, perhaps in hope that the students would quickly lose interest. But, as the protests persisted, and even grew, the stores began to assert their private property rights more

  2. #2 peak

    ays to look for toxicity in surface waters and sediments. Tests on ambient surface waters are conducted using the "water flea" (a crustacean), Ceriodaphnia dubia, to identify toxic effects on survival and reproduction. Collected

  3. #3 peak

    to a meteorite or bit of space junk that had entered the Earth’s atmosphere. As it turns out after further analysis, the explosion was due to an asteroid about 5

  4. #4 peak

    1920s, many hospitals in the United States had central libraries of patient information, with employees specifically hired to keep these records in good order. As time passed, their tasks became more complicated. The employees responsible for this work, who

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. L23:535 strongly excites
  2. L23:127 strongly suppresses
  3. L22:444 strongly suppresses
  4. L20:2032 strongly suppresses
  5. L19:2475 excites
  6. L23:108 suppresses
  7. L21:276 excites
  8. L19:1087 excites
  9. L21:1181 excites
  10. L21:958 suppresses
  11. L22:479 excites
  12. L22:2292 excites opening year or date range
  13. L21:3065 excites
  14. L15:720 excites
  15. L19:2536 excites
  16. L21:2148 suppresses
  17. L21:2103 excites
  18. L20:2672 excites
  19. L13:1926 suppresses
  20. L18:2303 excites
  21. L15:2655 suppresses
  22. L16:1532 excites
  23. L19:545 suppresses
  24. L20:1949 excites

Feeds later layers

  1. L25:349 strongly suppresses
  2. L25:2360 suppresses
  3. L27:1451 suppresses
  4. L27:382 suppresses
  5. L27:1281 suppresses
  6. L26:144 suppresses
  7. L27:217 suppresses
  8. L25:848 suppresses
  9. L26:138 suppresses
  10. L27:1110 suppresses
  11. L27:869 suppresses
  12. L26:1298 suppresses verb slot after a long subject
  13. L27:1295 suppresses
  14. L27:1032 suppresses
  15. L26:891 weakly suppresses
  16. L27:1596 weakly suppresses
  17. L25:1726 weakly excites
  18. L26:1716 weakly suppresses
  19. L26:1113 weakly suppresses
  20. L25:2156 weakly suppresses
  21. L25:1068 weakly suppresses
  22. L26:1188 weakly suppresses
  23. L26:1512 weakly suppresses
  24. L27:1222 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 -14.13 and -13.411 values between -12.68 and -11.951 values between -11.23 and -10.501 values between -10.50 and -9.771 values between -9.05 and -8.321 values between -8.32 and -7.594 values between -7.59 and -6.874 values between -6.87 and -6.145 values between -6.14 and -5.415 values between -5.41 and -4.697 values between -4.69 and -3.967 values between -3.96 and -3.239 values between -3.23 and -2.5121 values between -2.51 and -1.7842 values between -1.78 and -1.06187 values between -1.06 and -0.332570 values between -0.33 and 0.40983 values between 0.40 and 1.1329 values between 1.13 and 1.859 values between 1.85 and 2.586 values between 2.58 and 3.313 values between 3.31 and 4.033 values between 4.03 and 4.761 values between 4.76 and 5.482 values between 5.48 and 6.211 values between 8.39 and 9.122 values between 9.12 and 9.841 values between 13.48 and 14.201 values between 14.20 and 14.93offnegativepositiveoften

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. _throw
  2. era
  3. era
  4. inherit
  5. er
  6. ERA
  7. erin
  8. erland
  9. ER
  10. onn

Suppresses

  1. 进入了
  2. reversed
  3. 点了
  4. 给了
  5. 出来了
  6. 了一批
  7. exploded
  8. 学会了
  9. exited
  10. migrated