Neuron Explorer

Qwen3-0.6B

the letter B in vitamin names

A neuron in layer 10 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 letter "B" when it names a vitamin, like "vitamin B12", and also on "Bi" and "Biological" in journal names. Turning it up did not change the model's writing in any obvious way.

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

    from the outer layer of the wheat grain. The cereal has a rich content of protein, fiber, vitamins, minerals and phytonutrients. Vitamins include B1, B2, B3, B6 and E. Mineral content includes

  2. #2 peak

    level⏎of 52 pmol/L, whose mother had a level of 300 pmol/L. Understanding the clinical⏎manifestations of vitamin B12 deficiency and how it is relevant to Australia’s refugee⏎population might assist

  3. #3 peak

    .⏎Oats are one of the healthiest cereals on the planet. They are rich in healthy fiber, magnesium, manganese, thiamine (vitamin B1) and other important nutrients.⏎Protein content: 15% of

  4. #4 peak

    , we could:⏎|Vitamin B12 Need||94%||Confirm|⏎|Lupus (S

  5. #5 peak

    ini, L. (2021). Contrasting response of native and non‐native plants to disturbance and herbivory in mountain environments. Journal of Biogeography, 48(7), 1594-1

  6. #6 peak

    diets and the low vitamin B12 levels in this population.3⏎Vitamin B12 is one of the most complex vitamins. The measurement of serum levels and

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: the idea that a ‘absolute’ (i) is a non-unique (i) value. (i) is the same as the value of the (…) that can be filled in in the empty place where the value 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

    staffing gaps in order to provide students with instruction. The problem is not that schools do not wish to hire certified teachers but that a suitable pool of certified teachers qualified to teach in these fields and willing to

  2. #2 peak

    ways in which requirements influence designs and code (Figure 1).⏎From Requirements to Designs and Code⏎The boundary between requirements and design is not a sharp line, but keep your specifications free from implementation bias except when you have a compelling reason to intentionally

  3. #3 peak

    is somewhat common currency, but to claim tradition plays another role besides that of explaining how we came to have our moral beliefs, that it’s not only the medium through which these beliefs have been given to us, but also a guide for our present

  4. #4 peak

    recording of a human voice.⏎The recording played Thursday predates Thomas Edison's invention of the phonograph (previously thought to have recorded the first sound) by 17 years. It captured about 10 seconds of the French folks

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. L0:825 strongly suppresses
  2. L2:1869 excites
  3. L2:1855 excites
  4. L1:4 excites
  5. L4:1319 excites the 'B' in vitamin B
  6. L6:1357 excites
  7. L0:1638 suppresses
  8. L6:328 suppresses
  9. L7:520 suppresses
  10. L7:2095 excites
  11. L2:1807 weakly excites
  12. L0:1746 weakly excites
  13. L3:1733 weakly suppresses
  14. L4:667 weakly suppresses
  15. L2:1695 weakly excites
  16. L2:776 weakly suppresses
  17. L2:3060 weakly excites
  18. L7:1851 weakly excites
  19. L5:138 weakly excites
  20. L2:43 weakly suppresses
  21. L2:411 weakly excites
  22. L3:1254 weakly suppresses
  23. L2:1438 weakly suppresses
  24. L1:1810 weakly suppresses

Feeds later layers

  1. L18:385 strongly suppresses
  2. L18:242 strongly suppresses
  3. L20:256 strongly suppresses
  4. L18:402 strongly suppresses
  5. L19:370 excites
  6. L17:517 suppresses "but" in "not only X, but" phrases
  7. L19:583 suppresses
  8. L21:706 suppresses
  9. L19:2330 suppresses
  10. L22:2245 suppresses
  11. L22:1412 suppresses
  12. L19:1021 suppresses
  13. L23:2304 suppresses
  14. L26:2842 suppresses
  15. L26:2701 weakly suppresses
  16. L17:1721 weakly suppresses 'by' introducing a method or means
  17. L16:570 weakly suppresses 'but' in 'not only X, but also' contrasts
  18. L22:1533 weakly suppresses
  19. L26:2888 weakly suppresses
  20. L26:1364 weakly excites
  21. L26:1633 weakly suppresses
  22. L27:2209 weakly suppresses
  23. L22:1600 weakly suppresses
  24. L27:2670 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.

1 values between -2.06 and -1.971 values between -1.97 and -1.882 values between -1.80 and -1.711 values between -1.71 and -1.626 values between -1.53 and -1.444 values between -1.44 and -1.355 values between -1.35 and -1.2611 values between -1.26 and -1.177 values between -1.17 and -1.0817 values between -1.08 and -0.9923 values between -0.99 and -0.9027 values between -0.90 and -0.8157 values between -0.81 and -0.7263 values between -0.72 and -0.63102 values between -0.63 and -0.54145 values between -0.54 and -0.46163 values between -0.46 and -0.37245 values between -0.37 and -0.28301 values between -0.28 and -0.19339 values between -0.19 and -0.10414 values between -0.10 and -0.01442 values between -0.01 and 0.08389 values between 0.08 and 0.17360 values between 0.17 and 0.26281 values between 0.26 and 0.35218 values between 0.35 and 0.44138 values between 0.44 and 0.5390 values between 0.53 and 0.6234 values between 0.62 and 0.7112 values between 0.71 and 0.804 values between 0.80 and 0.893 values between 0.89 and 0.971 values between 1.06 and 1.151 values between 1.24 and 1.331 values between 1.42 and 1.51offnegativepositiveoften

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. streams
  2. wright
  3. ylene
  4. hores
  5. kor
  6. form
  7. 阄
  8. level
  9. lle
  10. _AST

Suppresses

  1. illance
  2. akedown
  3. serie
  4. ɥ
  5. cellphone
  6. series
  7. 再造
  8. 博物
  9. phones
  10. Strategic