Neuron Explorer

Qwen3-0.6B

"that" after demand or recommend verbs

A neuron in layer 3 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 fires on the word "that" when it follows verbs like demand, recommend, insist or order, as in "demanding that" or "recommended that". Turning it up did not clearly bring out this theme in the written text.

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

    atoil stakeholders, Folksam, a Swedish insurance company and KPA, a Swedish pension fund, have already sent in their votes backing a Greenpeace motion demanding that Statoil withdraw its investments.⏎Greenpeace is making the tar sands a key

  2. #2 peak

    irus. The CDC recommends that businesses encourage employees to get a seasonal flu vaccine and discourage sick employees from coming<|endoftext|>

  3. #3 peak

    k said. The American Indian Movement was founded in the late 1960s to protest the U.S. government’s treatment of Native Americans and demand that the government honor its treaties with Indian tribes.⏎He had a leading role in the

  4. #4 peak

    .⏎Amaziah lived during a time when a bitter dispute raged in Judah and Israel between the "Yahweh-only" party, which insisted that no other god but the Lord must be worshiped, and the more cosmopolitan

  5. #5 peak

    were circumcised by the same person, the report said. The New York City Commissioner of Health has issued a directive ordering that person to stop performing direct orogenital suction during circumcisions.⏎At least

  6. #6 peak

    What’s the point of genetic testing? It’s‘not actionable.’”⏎The American Academy of Pediatrics, along with other leading health professional organizations, has long recommended that all individuals with autism receive some type of genetic testing, such as screening for fragile

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: let me try to figure it out, and I am about the time being... I need to stop. So, if you are not a computer scientist, just to stop... I am not a real person, just to stop... I

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

    a miracle (the Incarnation) not as an end to itself, but as a means to allow us to be near Him in worship (through Christ’s redemptive sacrifice for our sins). Jesus did not come only<|endoftext|>WEDNESDAY

  2. #2 peak

    been a target of those who invaded India in order to spread Islam, with the might of their swords, from the beginning. When we read Muslim texts like Bahristan-i-Shahi, we can imagine the kind of barbarism unleashed on Kashmir

  3. #3 peak

    brain<|endoftext|>Harm Is in the Eye of the Beholder⏎Mark Davis is the DeWitt Wallace Professor and Chair of Biology at Macalester College, St. Paul,

  4. #4 peak

    , Philip Liebart. The design incorporates symbols of the three countries of the British Isles: A cross for England, a thistle for Scotland, and a shamrock for Ireland. The most notable jewel in the crown is a pale yellow brilliant in

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. L2:1036 weakly excites
  2. L1:773 weakly excites
  3. L1:1464 weakly excites
  4. L1:2823 weakly excites
  5. L2:75 weakly suppresses
  6. L0:874 weakly excites
  7. L2:746 weakly excites
  8. L1:1973 weakly suppresses
  9. L1:1029 weakly excites
  10. L2:41 weakly suppresses
  11. L2:2458 weakly excites
  12. L1:146 weakly excites
  13. L2:1469 weakly suppresses
  14. L0:2844 weakly excites
  15. L0:2255 weakly suppresses
  16. L1:2313 weakly suppresses
  17. L2:1710 weakly suppresses
  18. L2:2229 weakly suppresses
  19. L0:1301 weakly suppresses
  20. L2:1260 weakly excites
  21. L1:799 weakly excites
  22. L0:1224 weakly suppresses
  23. L0:1922 weakly suppresses
  24. L1:2890 weakly suppresses

Feeds later layers

  1. L24:1641 suppresses
  2. L7:1305 weakly excites
  3. L4:1521 weakly suppresses
  4. L22:924 weakly excites
  5. L20:2713 weakly excites
  6. L10:2491 weakly suppresses
  7. L27:895 weakly excites
  8. L13:2164 weakly suppresses
  9. L9:1709 weakly suppresses
  10. L5:2296 weakly excites
  11. L17:2418 weakly suppresses
  12. L11:2309 weakly suppresses
  13. L24:1089 weakly suppresses
  14. L4:1100 weakly suppresses
  15. L19:2989 weakly excites
  16. L8:1512 weakly suppresses
  17. L9:756 weakly excites
  18. L22:1553 weakly excites
  19. L24:1829 weakly suppresses
  20. L16:2117 weakly suppresses
  21. L23:167 weakly suppresses
  22. L11:85 weakly excites
  23. L6:2605 weakly suppresses
  24. L17:2046 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.

3 values between -0.68 and -0.5910 values between -0.59 and -0.4917 values between -0.49 and -0.4063 values between -0.40 and -0.30173 values between -0.30 and -0.21491 values between -0.21 and -0.111356 values between -0.11 and -0.021290 values between -0.02 and 0.08348 values between 0.08 and 0.1798 values between 0.17 and 0.2737 values between 0.27 and 0.378 values between 0.37 and 0.464 values between 0.46 and 0.565 values between 0.56 and 0.651 values between 0.65 and 0.751 values between 1.13 and 1.221 values between 1.41 and 1.512 values between 3.04 and 3.13offnegativepositiveoften

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. uries
  2. ileaks
  3. Driven
  4. }`).
  5. 床
  6. 蒜
  7. xec
  8. CBC
  9. policy
  10. 英

Suppresses

  1. ож
  2. bakım
  3. ertation
  4. Mormon
  5. KEN
  6. didn
  7. 拜师
  8. issions
  9. 阴影
  10. armor