Neuron Explorer

Qwen3-0.6B

'that' after reporting verbs

A neuron in layer 1 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 introduces a claim after verbs like 'reported', 'stated' or 'promised', as in 'the CDC also reports that'. Turning it up sent the model into vague sentences that did not clearly show this behaviour.

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

    by large bodies of cavalry and several regiments of infantry, while the heights on both sides were crowned by Russian artillery and infantry. Lord Raglan, perceiving that it was the intention of the Russians to carry off the guns they had captured from

  2. #2 peak

    pe climates, usually border desert climates. In both climate groups, the temperature change between day and night is considerable. One reason for the wide swings in temperature is that the skies are clear and the air is dry. Clouds would reflect much of

  3. #3 peak

    children tend to have more.⏎They estimate that more than 29 million Americans have a sinus infection or other sinus-related problems each ear.⏎The CDC also reports that influenza has resulted in nine to 45 million illnesses, 140

  4. #4 peak

    the size of a similar-size hoofed mammal and eyes only a third as large.⏎The combined effect was that Myotragus was sluggish, with slow reaction times, the bone study suggests

  5. #5 peak

    during this time, and that there was unequal access to maize between the sites, which the authors argue may have led to social tension through inequality.⏎The authors state that a combination of factors may have contributed to the high level of lethal trauma in the

  6. #6 peak

    in the Old Testament, faith was the only way the Saints could receive God’s promises. In the Garden of Eden, God promised that one day a Redeemer would come. Their hope was a future hope. They

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: I have been thinking about lately, and I have been thinking about lately, and I have been thinking about lately, and I have been thinking about lately, and I have been thinking about lately, and I have been thinking about lately, and

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

    003.⏎This article compares three eighteenth century evaluations about jesuitic missions among the Guaranis - Relação abbreviada da republica, que os religiosos jesuitas estabelecerão no Paraguai

  2. #2 peak

    grassy plateau that sweeps across the South African interior. The city bestrides the Witwatersrand, or Rand, a string of low, rocky ridges that constitutes the watershed between the

  3. #3 peak

    ANE?⏎Hurricanes are tropical cyclones that occur in the Atlantic Ocean and have winds with a sustained speed of at least

  4. #4 peak

    clothing sun protection,” by the Skin Cancer Foundation (1), UPF (Ultraviolet Protection Factor) treatment is designed to limit the amount of ultraviolet rays that penetrate your clothing’s fabric.⏎- SPF: On the other hand, SPF (

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. L0:1224 suppresses
  2. L0:178 suppresses
  3. L0:1301 suppresses
  4. L0:1966 weakly suppresses
  5. L0:1313 weakly excites
  6. L0:664 weakly excites
  7. L0:489 weakly excites
  8. L0:962 weakly suppresses
  9. L0:2744 weakly excites
  10. L0:113 weakly excites
  11. L0:1234 weakly excites
  12. L0:2084 weakly excites
  13. L0:470 weakly excites
  14. L0:556 weakly suppresses
  15. L0:1374 weakly suppresses
  16. L0:2841 weakly suppresses
  17. L0:2069 weakly excites
  18. L0:1729 weakly suppresses
  19. L0:1727 weakly excites
  20. L0:1210 weakly suppresses
  21. L0:84 weakly suppresses
  22. L0:2827 weakly suppresses
  23. L0:2204 weakly excites
  24. L0:574 weakly excites

Feeds later layers

  1. L17:1361 suppresses 'that' after verbs like recommend
  2. L4:2517 weakly suppresses
  3. L3:1485 weakly suppresses
  4. L21:571 weakly excites
  5. L3:1418 weakly excites
  6. L3:905 weakly suppresses
  7. L5:953 weakly excites
  8. L4:956 weakly suppresses 'that' after a noun, introducing a clause
  9. L11:1524 weakly excites
  10. L8:2274 weakly excites
  11. L26:963 weakly suppresses
  12. L18:160 weakly suppresses
  13. L7:2551 weakly excites
  14. L19:1703 weakly suppresses
  15. L13:344 weakly suppresses
  16. L2:45 weakly excites
  17. L12:1655 weakly excites
  18. L19:869 weakly suppresses
  19. L23:3019 weakly excites
  20. L6:665 weakly suppresses
  21. L23:743 weakly excites
  22. L26:106 weakly suppresses
  23. L13:2049 weakly excites
  24. L19:2174 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 -4.40 and -4.251 values between -3.82 and -3.681 values between -3.68 and -3.543 values between -3.40 and -3.251 values between -3.11 and -2.971 values between -2.68 and -2.541 values between -2.40 and -2.262 values between -2.26 and -2.111 values between -1.83 and -1.693 values between -1.69 and -1.543 values between -1.40 and -1.262 values between -1.26 and -1.123 values between -0.97 and -0.831 values between -0.83 and -0.692 values between -0.54 and -0.4015 values between -0.40 and -0.26142 values between -0.26 and -0.122064 values between -0.12 and 0.031516 values between 0.03 and 0.17124 values between 0.17 and 0.3113 values between 0.31 and 0.455 values between 0.45 and 0.601 values between 0.60 and 0.741 values between 0.88 and 1.021 values between 1.17 and 1.31offnegativepositiveoften

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. stakes
  2. 这种情况
  3. ellation
  4. uforia
  5. 帐
  6. iframe
  7. account
  8. 五个
  9. 童话
  10. spoilers

Suppresses

  1. eness
  2. onn
  3. een
  4. bane
  5. 税务局
  6. uetype
  7. nak
  8. erne
  9. né
  10. developed