Neuron Explorer

Qwen3-0.6B

"or" between alternatives

A neuron in layer 12 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 "or" when it offers a choice between two options, such as "fat-free or low-fat" or "signs or symptoms", especially in health and diet writing. Turning it up did not produce text that clearly showed this 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

    your doctor.⏎Healthy Weight Tip⏎Set goals to eat healthy like replacing full-fat version of foods such as milk, yogurt, and salad dressing with the fat-free or low-fat kind.⏎The best way to give your body the balanced nutrition it needs

  2. #2 peak

    hormone, T4 (thyroxine) and T3 (triiodothyronine).⏎Many babies with congenital hypothyroidism appear normal at birth or for several months after birth. But others may have these signs and symptoms:⏎They

  3. #3 peak

    The world is filled with various fad diets. From keto diet to paleo diet and atkin diet - almost all of these weight loss diets recommend omitting or reducing carbs from your daily meal plan.⏎How Low-Carb Diet

  4. #4 peak

    -4 tablespoons of canned organic pumpkin (just the pumpkin puree – because pumpkin pie filling has lots of added sugar)⏎• Two tablespoons of Greek yogurt – plain or honey flavor (it’s higher in protein than other yogurt); or a non-d

  5. #5 peak

    as beans, lentils, chickpeas, black-eyed peas; vegetables, such as broccoli, cabbage, and carrots. Heart-healthy eating includes fat-free or low-fat dairy products, such as, skim milk, fish high in omega-

  6. #6 peak

    ERP systems, organizations are still being relegated to the use of manual, paper-oriented processes and forms to fulfill various core business functions that their ERP systems do not provide or do not provide correctly. According the Gartner Groups, as much as 8

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: what if it is the same as a person or another person, what if it is the same as a person, what if it is the same as a person, what if it is the same as a person, what if it is the

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 autonomous aircraft will snapshot aerial views of Martian cliffs, caves and craters that the land-bound rover can’t explore. And even if it doesn’t take flight, the rover can still gather important data from the surface.⏎NASA said it

  2. #2 peak

    is legally valid, de facto polygamy is generally permitted, as long as adultery is not illegal. In areas like these, where polygamy is prohibited but tolerated, there is no legal recognition for more spouses after the first. In an

  3. #3 peak

    the serum was tested for antibodies to the virus with a standard antibody test. Although no live virus was recovered and antibodies to avian influenza virus could not be detected, four specimens

  4. #4 peak

    this way, Bradbury is able to evoke emotion in the reader, the mark of a successful narrative. Thus, although there are no humans within the story itself, the audience continues to see the effects of their presence long after they

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. L11:2994 weakly suppresses
  2. L3:1513 weakly suppresses
  3. L7:662 weakly excites
  4. L2:2430 weakly excites
  5. L7:1171 weakly excites
  6. L6:1573 weakly suppresses
  7. L2:503 weakly excites
  8. L7:2208 weakly suppresses
  9. L10:2215 weakly suppresses
  10. L0:2659 weakly excites
  11. L7:2318 weakly excites
  12. L6:1523 weakly suppresses
  13. L2:1224 weakly suppresses
  14. L4:1043 weakly suppresses
  15. L11:871 weakly excites zero and infinity as values
  16. L8:487 weakly excites
  17. L10:2041 weakly suppresses
  18. L11:2468 weakly excites
  19. L9:1191 weakly suppresses sentence-ending full stops in web text
  20. L11:895 weakly excites
  21. L7:1086 weakly excites
  22. L9:540 weakly excites
  23. L6:1868 weakly excites
  24. L8:2354 weakly suppresses

Feeds later layers

  1. L15:2800 strongly excites
  2. L13:2892 suppresses
  3. L13:791 suppresses
  4. L16:950 suppresses
  5. L14:2547 excites
  6. L16:998 weakly excites
  7. L13:919 weakly suppresses
  8. L18:1892 weakly suppresses
  9. L18:623 weakly suppresses
  10. L16:2492 weakly suppresses
  11. L19:2060 weakly suppresses
  12. L13:705 weakly suppresses
  13. L19:1191 weakly suppresses
  14. L15:1573 weakly suppresses
  15. L17:2113 weakly suppresses
  16. L16:2709 weakly suppresses
  17. L13:2553 weakly excites
  18. L16:1570 weakly suppresses
  19. L14:1517 weakly excites
  20. L15:2216 weakly suppresses
  21. L14:2520 weakly suppresses
  22. L16:1634 weakly excites
  23. L19:2353 weakly suppresses
  24. L23:257 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.

2 values between -0.98 and -0.912 values between -0.91 and -0.844 values between -0.84 and -0.772 values between -0.77 and -0.708 values between -0.63 and -0.563 values between -0.56 and -0.496 values between -0.49 and -0.4218 values between -0.42 and -0.3551 values between -0.35 and -0.28122 values between -0.28 and -0.21256 values between -0.21 and -0.14543 values between -0.14 and -0.07802 values between -0.07 and -0.00923 values between -0.00 and 0.07547 values between 0.07 and 0.14344 values between 0.14 and 0.21169 values between 0.21 and 0.2863 values between 0.28 and 0.3527 values between 0.35 and 0.428 values between 0.42 and 0.493 values between 0.49 and 0.562 values between 0.56 and 0.631 values between 0.77 and 0.841 values between 1.33 and 1.401 values between 1.75 and 1.82offnegativepositiveoften

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. 💛
  2. Leone
  3. 平凡
  4. justice
  5. ��
  6. plusplus
  7. ịnh
  8. 凰
  9. _SZ
  10. scaleX

Suppresses

  1. consequently
  2. uada
  3. quired
  4. aven
  5. trie
  6. ◻
  7. Aim
  8. yet
  9. Consequently
  10. ctica