Neuron Explorer

Qwen3-0.6B

the word 'directly'

A neuron in layer 7 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 "directly" in phrases like "convert it directly" or "purchased directly from." When it was turned up, the model's writing often slipped in the word "directly" too, as in "the sun was shining directly over the ocean." It goes quiet, or slightly negative, on phrases about right angles.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    .⏎A Comparison Of High-Level And Low-Level Programming Languages⏎A way to categorize programming languages is through high-level or low-level. The low-level languages communicate directly with the central processing unit or the CPU. They can perform very basic commands but

  2. #2 peak

    from DC to AC, these new solar panels have a⏎solar micro inverter that takes the DC current generated by the⏎solar panel and converts it directly to AC current before leaving⏎the panel.<|endoftext|>When discussing air quality, TV

  3. #3 peak

    . “The EM Drive (Electro Magnetic Drive) uses electromagnetic microwave cavities to directly convert electrical energy to thrust without the need to expel any propellant,” explains

  4. #4 peak

    or a blend of both, by choice of catalyst). Liquid carbon dioxide is also produced by the process, this is sold to industry but in future could be fed directly into a carbon capture and storage system.⏎The success of these pilots led directly to

  5. #5 peak

    sensors to study how vegetation in the region transports water from the soil into the atmosphere. The lasers allow the researchers to determine how much water in the air evaporates directly from the soil and how much evaporates from the leaves of plants. “We

  6. #6 peak

    or disturbed. You could be exposed to mould if you breathe in mould spores, breathe in the toxins that certain moulds produce, swallow mouldy material or directly handle mouldy material.⏎People who work in certain occupations may have higher exposures to

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 to have fun with this problem. This is all about how it's going to be used. So, if you really, really like it. (for example, maybe something very simple or something very fun or something very strange)

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

    triangle is equal to the sum of the squares of the sides containing the right angle." But Pythagoras had an all-inclusive vision, which also included the

  2. #2 peak

    the angle right is a little more complicated.⏎In order to better study this use of tails,

  3. #3 peak

    X is a moveable point on the hypotenuse, and P and Q are the feet⏎of the perpendiculars from X to the sides of a right angled⏎triangle. What position of X makes the length of PQ a minimum?⏎The

  4. #4 peak

    you is spinning—the cause is often a vestibular (inner ear) disorder known as benign paroxysmal positional vertigo (BPPV).⏎BPPV is one of the most common types of

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. L3:2253 excites
  2. L4:1751 excites
  3. L5:2827 weakly suppresses
  4. L5:2465 weakly excites
  5. L3:2575 weakly excites
  6. L2:75 weakly suppresses
  7. L6:2778 weakly suppresses
  8. L0:2924 weakly suppresses
  9. L6:2440 weakly excites
  10. L0:1835 weakly suppresses
  11. L3:2291 weakly excites
  12. L6:1722 weakly excites
  13. L1:2683 weakly excites
  14. L3:2349 weakly excites
  15. L0:1602 weakly suppresses
  16. L4:1981 weakly excites
  17. L3:2437 weakly excites
  18. L6:2393 weakly excites
  19. L3:1866 weakly excites
  20. L2:1850 weakly suppresses
  21. L2:2708 weakly excites
  22. L2:1913 weakly suppresses unclear: word closing a two-word term
  23. L5:500 weakly excites
  24. L6:1939 weakly excites

Feeds later layers

  1. L8:1569 suppresses
  2. L20:1976 weakly suppresses
  3. L10:1708 weakly excites
  4. L15:1399 weakly excites
  5. L21:2885 weakly suppresses
  6. L11:541 weakly excites
  7. L10:1949 weakly excites
  8. L22:1070 weakly excites
  9. L11:1364 weakly suppresses
  10. L11:1429 weakly suppresses
  11. L15:2938 weakly suppresses comma after an if-clause
  12. L8:3016 weakly excites
  13. L18:1099 weakly suppresses
  14. L11:2932 weakly suppresses
  15. L17:1556 weakly excites 'instead of' and 'rather than'
  16. L22:1335 weakly suppresses pair and compound words in technical text
  17. L24:2556 weakly suppresses
  18. L8:849 weakly excites
  19. L9:970 weakly suppresses
  20. L14:877 weakly suppresses
  21. L11:685 weakly excites
  22. L10:1016 weakly suppresses
  23. L9:1149 weakly suppresses
  24. L26:2926 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 -1.25 and -1.131 values between -1.13 and -1.011 values between -1.01 and -0.892 values between -0.89 and -0.781 values between -0.78 and -0.661 values between -0.66 and -0.5411 values between -0.54 and -0.4259 values between -0.42 and -0.31267 values between -0.31 and -0.19742 values between -0.19 and -0.071647 values between -0.07 and 0.05862 values between 0.05 and 0.17238 values between 0.17 and 0.2852 values between 0.28 and 0.407 values between 0.40 and 0.522 values between 0.52 and 0.643 values between 0.64 and 0.753 values between 0.75 and 0.871 values between 0.87 and 0.992 values between 0.99 and 1.111 values between 1.11 and 1.222 values between 1.22 and 1.341 values between 2.52 and 2.641 values between 3.34 and 3.46offnegativepositiveoften

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. Threads
  2. utherford
  3. imations
  4. threads
  5. innen
  6. iffs
  7. olución
  8. [min
  9. 硬件
  10. 诔

Suppresses

  1. сон
  2. -mean
  3. 笛
  4. tim
  5. dur
  6. звуч
  7. constant
  8. еча
  9. CONSTANT
  10. -Ass