Neuron Explorer

Qwen3-0.6B

'that' after verbs like recommend

A neuron in layer 17 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 responds to the word 'that' when it introduces a clause after words like 'recommend', 'demand' or 'require'. The opposite side fires on 'that' after a noun, as in 'organizations that'. Turning it up gave repetitive text with no clear 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

    to share views with audience⏎Establish eye contact with the person⏎Direct questions at the person or someone near the person⏎Introduce a time for thinking and ask that everyone be silent for a moment. Walk near that person. Empower others to

  2. #2 peak

    , animal cruelty convictions are classified as either a misdemeanor or a felony. If a person under the age of eighteen is convicted of animal cruelty, the law also requires that he or she undergo counseling. On the other hand, if a person is brought

  3. #3 peak

    see local events and offers. Based on your location, we recommend that you select: .⏎You can also select a web site from the following list:⏎

  4. #4 peak

    Based on your location, we recommend that you select: .⏎You can also select a web site from the following list:⏎

  5. #5 peak

    't be given garlic.⏎- Dr's Foster and Smith of peteducation.com are in agreement that garlic causes a breakdown of red blood cells in dogs though they permit that garlic in terribly tiny amounts in some industrial pet foods has not been shown to cause

  6. #6 peak

    and made friends with a number of reformers who crossed his path.⏎Uncle Laski did not like this at all. He ordered John home and demanded that he recant his "heretical" beliefs. John obeyed and was made

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: is is is, is is, is, is, is, is: repeats is is - is, repeats is - is - (is, repeats is, is, repeats - (is, repeats: is, repeats: repeats: is

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

    The journal has been published since 1964. AIBS is a meta-level organization for professional scientific societies and organizations that are involved with biology. It represents nearly 160 member societies and organizations

  2. #2 peak

    water doesn't get cold enough. Tree-harming insects in the Piedmont and mountains are also thriving in milder winters. Atlantic surf clams that fisherman used to catch as far south as Cape Hatteras are now rarely

  3. #3 peak

    caching in clients) and the Control and View (possibly combined) reside within the client.⏎There are also applications where we would like to subcribe to dynamic content that is continuously pushed from the

  4. #4 peak

    5 hours at a time.⏎They have 4 lung chambers, so you will usually find them coming up to the surface 4 times to take a breath that fills each chamber. As the turtles move their flippers, it pumps the

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. L16:1303 excites 'that' starting a clause after a noun
  2. L9:661 excites
  3. L13:344 suppresses
  4. L3:1485 excites
  5. L6:2111 excites "that" starting a relative clause
  6. L1:1562 suppresses 'that' after reporting verbs
  7. L9:865 suppresses
  8. L16:671 excites
  9. L8:1001 weakly suppresses
  10. L2:126 weakly excites
  11. L3:59 weakly excites
  12. L4:596 weakly suppresses
  13. L3:1003 weakly excites
  14. L8:10 weakly excites
  15. L4:12 weakly suppresses
  16. L8:16 weakly suppresses
  17. L4:2517 weakly suppresses
  18. L15:78 weakly excites
  19. L6:13 weakly excites
  20. L8:7 weakly excites
  21. L15:210 weakly excites
  22. L16:109 weakly excites
  23. L9:75 weakly excites 'what' or 'how' after see or understand
  24. L15:39 weakly suppresses

Feeds later layers

  1. L21:2475 suppresses "that" starting a relative clause
  2. L21:123 suppresses
  3. L18:391 suppresses
  4. L18:1246 suppresses
  5. L23:220 suppresses
  6. L26:669 suppresses
  7. L20:349 excites
  8. L19:2332 suppresses
  9. L18:1985 suppresses
  10. L19:617 suppresses
  11. L18:452 suppresses
  12. L18:715 suppresses
  13. L20:586 suppresses
  14. L20:241 suppresses
  15. L19:1703 suppresses
  16. L19:168 suppresses
  17. L18:2811 suppresses
  18. L23:127 excites
  19. L19:2335 suppresses
  20. L26:436 suppresses
  21. L21:212 excites
  22. L22:518 suppresses
  23. L19:1972 suppresses
  24. L19:1317 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 -52.53 and -51.131 values between -46.93 and -45.541 values between -44.14 and -42.743 values between -41.34 and -39.941 values between -39.94 and -38.541 values between -38.54 and -37.141 values between -37.14 and -35.743 values between -35.74 and -34.344 values between -32.94 and -31.541 values between -31.54 and -30.142 values between -30.14 and -28.751 values between -27.35 and -25.951 values between -24.55 and -23.152 values between -21.75 and -20.351 values between -20.35 and -18.951 values between -18.95 and -17.553 values between -16.15 and -14.752 values between -13.35 and -11.951 values between -11.95 and -10.565 values between -9.16 and -7.761 values between -7.76 and -6.362 values between -6.36 and -4.964 values between -4.96 and -3.5619 values between -3.56 and -2.1648 values between -2.16 and -0.762856 values between -0.76 and 0.64940 values between 0.64 and 2.042 values between 2.04 and 3.44offnegativepositiveoften

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. wró
  2. yes
  3. ử
  4. leur
  5. minded
  6. 完
  7. typeof
  8. itized
  9. 里程碑
  10. 那儿

Suppresses

  1. aits
  2. �
  3. includes
  4. imates
  5. operates
  6. .activities
  7. incorporates
  8. projects
  9. isters
  10. allows