Neuron Explorer

Qwen3-0.6B

"even" in concessive phrases

A neuron in layer 15 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 "even" when it introduces a surprising condition, as in "even if", "even when" or "even in small amounts". It goes quiet, or slightly negative, on words like "do", "simple" and "enough" in phrases such as "will do the job". When turned up, the model's writing drifted toward "without" and "no need" phrases, which matches what it promotes.

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

    weight, reduce susceptibility to infection, and increase cognitive function in the short run. Because school meals are usually fortified, a child's micronutrient intake can improve even if her total calorie consumption does not. These xi short-run gains may improve a

  2. #2 peak

    to defend their territory or pack members from the stranger in the house.⏎The fourth type of anxiety in dogs is generalized anxiety. Some dogs can exhibit signs of anxiety even when there seems nothing to be worried about. Some vets believe

  3. #3 peak

    before eventually making their way into water sources. They can also arise from manure and waste runoff from septic tanks and sewer systems. Nitrate contamination is dangerous even in instances of short-term exposure. Children and infants can develop methemoglobinemia

  4. #4 peak

    amounts of cadmium, a known heavy metal that is toxic even in small concentrations. Remembering that anodes are sacrificial, that is they

  5. #5 peak

    not by families but colors.⏎Butterflies perform an important role in pollinating plants and they themselves are ecological indicators. “Unfortunately, their numbers are dwindling even in greener pockets of the city because of the use of pesticides and the we

  6. #6 peak

    because the appendix becomes inflamed and starts to put pressure on other organs in the abdomen, including the stomach. This can lead to a feeling of fullness, even if you haven't eaten much, and can make you lose your appetite.⏎N

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 working on this project, and I have been working on this project, and I have been working on this project... and I have been working on this project, and I have been working on this project... and I have been

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

    greening your cooking and kitchen:⏎• Limit the time the stove/oven is used. Never light the oven or turn on a burner when a small appliance will do the job. Microwave ovens, toaster ovens, electric griddles, pan

  2. #2 peak

    display immediately.⏎- To see the console display, you must run without debugging.⏎- The problem using Visual Studio is that many files will be created even for a simple program. All files are created inside the project folder.⏎- Under the project name

  3. #3 peak

    need for heat sinks largely disappears, eliminating the need for design tricks like thermal via’s or creating any sort of complicated thermal pathways. The electrical traces themselves are often enough to suitably cool small mid-power LED packages. Plus PCB’s are costly to

  4. #4 peak

    . Debs by Cassie Cook⏎1: Eugene V. Debs was born on November 5th 1855, Eugene left school early to become a railroad painter. During those years was the first time he became in tune

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. L13:593 excites
  2. L12:2992 suppresses
  3. L13:2225 excites
  4. L14:2990 excites
  5. L14:2237 suppresses
  6. L12:3057 excites
  7. L11:2712 suppresses
  8. L10:660 excites
  9. L13:1744 suppresses
  10. L12:2426 excites
  11. L12:1110 excites
  12. L13:389 excites
  13. L14:445 excites
  14. L13:528 suppresses
  15. L14:407 excites
  16. L11:2776 excites
  17. L12:1834 suppresses
  18. L14:1414 excites
  19. L13:2656 suppresses
  20. L11:554 suppresses
  21. L14:980 suppresses
  22. L14:1809 suppresses
  23. L14:536 suppresses
  24. L14:3026 suppresses

Feeds later layers

  1. L21:2710 excites
  2. L23:434 excites
  3. L20:2264 excites
  4. L17:2987 suppresses
  5. L17:1926 excites
  6. L20:2275 excites
  7. L16:1282 excites
  8. L16:1930 excites
  9. L17:2611 excites
  10. L26:2350 excites
  11. L18:2614 excites
  12. L20:2494 excites
  13. L18:2514 excites
  14. L18:1571 excites
  15. L26:406 excites
  16. L24:334 excites
  17. L22:2000 excites
  18. L18:2563 excites
  19. L17:2628 excites
  20. L20:2610 excites
  21. L17:2351 excites
  22. L24:1794 excites
  23. L18:2298 excites
  24. L19:2011 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 -2.00 and -1.853 values between -1.85 and -1.702 values between -1.70 and -1.561 values between -1.56 and -1.413 values between -1.41 and -1.263 values between -1.26 and -1.116 values between -1.11 and -0.967 values between -0.96 and -0.8114 values between -0.81 and -0.6628 values between -0.66 and -0.5144 values between -0.51 and -0.36141 values between -0.36 and -0.21659 values between -0.21 and -0.062275 values between -0.06 and 0.09497 values between 0.09 and 0.24125 values between 0.24 and 0.3836 values between 0.38 and 0.5320 values between 0.53 and 0.6811 values between 0.68 and 0.837 values between 0.83 and 0.984 values between 0.98 and 1.135 values between 1.13 and 1.282 values between 1.28 and 1.435 values between 1.43 and 1.583 values between 1.58 and 1.732 values between 1.73 and 1.881 values between 2.02 and 2.171 values between 2.32 and 2.471 values between 2.62 and 2.771 values between 3.81 and 3.96offnegativepositiveoften

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. without
  2. without
  3. 无需
  4. WITHOUT
  5. 不需要
  6. без
  7. zonder
  8. WITHOUT
  9. Without
  10. 不必

Suppresses

  1. 严重
  2. (full
  3. 严重的
  4. 嚴重
  5. full
  6. 十足
  7. fullname
  8. nặng
  9. full
  10. 踵