Neuron Explorer

Qwen3-0.6B

'what' or 'how' after see or understand

A neuron in layer 9 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 'what' (or 'how') right after verbs like 'see', 'understand', or 'work out', where a question-like clause is about to follow. Turning it up did not bring out this theme in the sample text, which went off topic.

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

    and has not in fact used measurements of the Antarctic "hole". There's many a slip twixt cup and lip there. It will be interesting to see what is in the relevant journal article when -- and if -- it has passed peer review

  2. #2 peak

    ; let’s say we want to see what has changed between Commit 1 and Commit 3.⏎The diff will compare the

  3. #3 peak

    sight or at the very least supplementing what they see with extra-sensory information. For example, whales or bats,

  4. #4 peak

    lie completely within the designated 15-mile diameter count circle, and the site has opened its land up to birders since 1996 to see what interesting and unique birds call the site home or have stopped by on their way south

  5. #5 peak

    Level 5.0 - 6..5⏎Students will truly understand what they read with the help of these easytouse reading comprehension cards! Your students

  6. #6 peak

    stopped them. In evidencing dictionary items, OED editors are limited by and large to what’s available in print. This makes finding early examples of colloquial language very

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 don" 1. 1.1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1

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

    help. The sooner you receive care for your anxiety symptoms, the better. Feel free to reach out

  2. #2 peak

    majority of Company B, to whose leadership Little Wound was promoted as first sergeant. Major chiefs Red Cloud, Young

  3. #3 peak

    a heart-healthy lifestyle. Read more in these topics about reaching and staying at a healthy weight:⏎Seeing your doctor regularly and getting screening tests is important. The sooner your doctor diagnoses a disease, the more likely

  4. #4 peak

    "Maintaining and improving the chemical, biological, and physical integrity of the Great Lakes is the central commitment between the United States and Canada. The report documents how far we have come and how much further we need to go," said Dereth Gl

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. L4:927 excites
  2. L6:502 suppresses
  3. L4:470 suppresses
  4. L4:227 suppresses
  5. L7:2058 excites
  6. L2:1957 suppresses
  7. L4:2197 suppresses
  8. L4:981 suppresses
  9. L5:458 excites
  10. L8:132 suppresses
  11. L8:369 suppresses 'of' after 'learned' or 'because'
  12. L3:2122 excites
  13. L8:1997 excites
  14. L1:2889 suppresses
  15. L1:1457 weakly suppresses
  16. L8:2146 weakly excites "as" in "it seems as if" phrases
  17. L4:2719 weakly suppresses
  18. L3:1538 weakly excites 'what' as a free relative
  19. L4:1582 weakly excites
  20. L5:570 weakly excites
  21. L4:534 weakly suppresses
  22. L4:956 weakly excites 'that' after a noun, introducing a clause
  23. L1:1616 weakly suppresses
  24. L8:7 weakly excites

Feeds later layers

  1. L13:344 excites
  2. L16:2975 excites
  3. L17:1282 excites
  4. L17:1407 excites "and" joining questions (wh-question context)
  5. L16:817 excites
  6. L17:2368 excites
  7. L12:434 excites
  8. L16:948 excites
  9. L18:1617 excites
  10. L11:2591 excites
  11. L12:1667 excites
  12. L19:1703 excites
  13. L15:1473 excites
  14. L19:2737 excites
  15. L17:1884 excites
  16. L13:2728 excites
  17. L17:2967 excites
  18. L15:2893 excites
  19. L16:1078 excites
  20. L17:2280 excites
  21. L15:467 excites
  22. L21:639 weakly excites
  23. L16:528 weakly excites
  24. L17:2718 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 -3.43 and -3.091 values between -1.40 and -1.073 values between -1.07 and -0.73874 values between -0.73 and -0.392462 values between -0.39 and -0.06337 values between -0.06 and 0.2896 values between 0.28 and 0.6246 values between 0.62 and 0.9620 values between 0.96 and 1.2919 values between 1.29 and 1.637 values between 1.63 and 1.976 values between 1.97 and 2.315 values between 2.31 and 2.643 values between 2.64 and 2.982 values between 2.98 and 3.321 values between 3.32 and 3.662 values between 3.66 and 3.991 values between 3.99 and 4.331 values between 4.33 and 4.672 values between 4.67 and 5.001 values between 5.00 and 5.341 values between 5.34 and 5.683 values between 5.68 and 6.023 values between 6.02 and 6.352 values between 6.35 and 6.691 values between 6.69 and 7.031 values between 7.03 and 7.362 values between 7.36 and 7.701 values between 8.04 and 8.382 values between 9.39 and 9.732 values between 9.73 and 10.06offnegativepositiveoften

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. arsi
  2. eve
  3. 记者了解
  4. arp
  5. alam
  6. arra
  7. ever
  8. 我说
  9. briefed
  10. STDCALL

Suppresses

  1. ộng
  2. زي
  3. hài
  4. 过
  5. StateException
  6. Including
  7. 植物
  8. licer
  9. inconsistent
  10. 砚