Neuron Explorer

Qwen3-0.6B

'the' after words like 'underscores'

A neuron in layer 21 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 'the' right after verbs like 'underscores', 'emphasizes' or 'highlights', where the next word is likely to be 'importance' or 'need'. When it was turned up, the model kept producing 'the importance of the' over and over.

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

    in 2013 was 25 percent, or 13 percentage points below the state average. This means that after decades of being told about the need to recycle, we are one-third of the way to a goal we

  2. #2 peak

    children’s range of movement has shrunk over the decades and nowadays a child does not necessarily leave the house or its yard without the company of parents. This emphasizes the importance of the immediate environment from the front door.⏎With planning, we want to

  3. #3 peak

    mismatched interfaces. Layer inclination further influences stress fields, shifting the location of maximum stress concentration over varying distances. This study underscores the significance of accurately understanding mechanical properties, layer dip in volcanoes, and magma chamber

  4. #4 peak

    financial and social burden of NCDs will grow dramatically in the next years. The results highlight the need for interventions to address behavioral changes, including reduction in smoking, improvement of dietary

  5. #5 peak

    his professional and personal lives collided when he was ordered to evacuate his home in the middle of the night to escape a spreading wildfire. Lempert's experience emphasized the challenges of adapting to climate change, “not merely because it is hard, but

  6. #6 peak

    drought and floods, underscoring the connections between climate and people,” this article says. Yet the extensive civilization at Ang

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: how if we are all made to be in a way that if we do something that makes us act in a way that if we do something that makes us act in a way that if we do something that makes us act in a way that

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

    1925–1926.⏎[I’m a big fan of the excavation technique reproduced here.]⏎[Legalized pot hunting is more to the point.]⏎I’ve recently come under fire, again, for

  2. #2 peak

    , thematic survey of the history of science from its roots in different cultures around the world through to the present day. He explores crucial milestones in scientific development and at the same time examines the enormous social and intellectual changes they initiated. Opening with a discussion

  3. #3 peak

    computers connected to any network in the set. With the Internet, many remote computers can communicate.⏎Networks in Everyday Life⏎Extensive use of networks is on the business side. Networks are used for advertising, production, posting, editing, payment

  4. #4 peak

    understandable. Gravitational waves are minuscule — the atomic jitters that pass through our world when two black holes bash together in a distant galaxy are on the order of a billionth of a billionth the diameter of an atom. L

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. L20:1016 excites
  2. L19:1869 suppresses
  3. L16:419 excites
  4. L9:24 excites "the" before a named organisation or title
  5. L19:186 excites
  6. L18:2398 suppresses
  7. L10:1377 suppresses
  8. L16:792 weakly excites "the" inside a prepositional phrase
  9. L20:361 weakly excites
  10. L17:2270 weakly excites
  11. L20:2278 weakly suppresses
  12. L18:2809 weakly suppresses
  13. L13:1164 weakly suppresses
  14. L19:2690 weakly suppresses
  15. L17:611 weakly suppresses
  16. L19:72 weakly excites
  17. L0:400 weakly excites
  18. L19:419 weakly excites
  19. L18:641 weakly suppresses
  20. L7:2667 weakly suppresses
  21. L6:911 weakly suppresses
  22. L1:2193 weakly excites
  23. L17:818 weakly suppresses
  24. L19:2988 weakly excites

Feeds later layers

  1. L27:2019 excites
  2. L27:1550 excites
  3. L26:507 excites
  4. L23:1323 excites
  5. L22:691 excites
  6. L24:1062 excites
  7. L27:127 suppresses
  8. L22:3060 excites
  9. L27:399 excites
  10. L26:2119 excites
  11. L22:438 excites
  12. L23:2310 weakly excites
  13. L23:1508 weakly excites
  14. L26:1229 weakly excites
  15. L22:2782 weakly excites
  16. L22:2814 weakly excites
  17. L22:3036 weakly excites
  18. L23:1726 weakly excites
  19. L24:1456 weakly excites
  20. L22:2722 weakly suppresses
  21. L25:3049 weakly excites
  22. L22:2587 weakly excites
  23. L23:791 weakly excites
  24. L22:2387 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.

3 values between -15.24 and -14.031 values between -14.03 and -12.811 values between -12.81 and -11.591 values between -11.59 and -10.385 values between -10.38 and -9.163 values between -9.16 and -7.953 values between -7.95 and -6.7312 values between -6.73 and -5.5110 values between -5.51 and -4.3016 values between -4.30 and -3.0836 values between -3.08 and -1.8689 values between -1.86 and -0.653231 values between -0.65 and 0.57464 values between 0.57 and 1.7913 values between 1.79 and 3.004 values between 3.00 and 4.224 values between 4.22 and 5.436 values between 5.43 and 6.653 values between 6.65 and 7.871 values between 9.08 and 10.301 values between 13.95 and 15.161 values between 32.19 and 33.41offnegativepositiveoften

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. importance
  2. 重要
  3. 的重要性
  4. import
  5. Import
  6. Importance
  7. import
  8. Import
  9. importância
  10. 的重要

Suppresses

  1. stackpath
  2. achable
  3. ��
  4. etically
  5. astically
  6. eson
  7. {}".
  8. usra
  9. edly
  10. iffies