Neuron Explorer

Qwen3-0.6B

'that' after 'fact' or 'considering'

A neuron in layer 1 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 reacts to the word 'that' in phrases like 'due to the fact that' or 'considering that', where a reason is about to be explained. Its opposite side fires weakly on 'since' in 'ever since'. Turning it up did not produce a clear visible change in the writing.

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

    While this disability qualification is important for the fair protection of workers with COVID-induced disabilities, the link between COVID-19 and intellectual disabilities remains underrecognized.⏎Considering that the second-highest risk factor for COVID-19 diagnoses and deaths is intellectual

  2. #2 peak

    chain and works directly with the final buyer. Due to the fact that there are a lot of goods and services, and their specifics differ significantly, the

  3. #3 peak

    Everyone wants to live in a peaceful community free from conflict. Is it conceivable? Yes, if everyone chooses to get along with one another.⏎Due to the fact that the older and younger generations were not raised in the same manner, there are instances

  4. #4 peak

    symmetry of nature. For example, the conservation of energy follows from the time-invariance of physical systems, and the conservation of angular momentum arises from the fact that physical systems behave the same regardless of how they are oriented in space.⏎A partial

  5. #5 peak

    trial detention centers when they are arrested or in juvenile correctional facilities when they are adjudicated. This distinction is further complicated, as discussed later, by the fact that not all youth in juvenile justice facilities are necessarily delinquent.⏎Epidemi

  6. #6 peak

    to discuss the methods of the Cunninghams which they were happy farmers who refuse charity due to the fact that they couldn’t afford it. 2 The expert individuals is poor at that time

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 thinking about myself, and I have been thinking about myself. And I have been thinking about myself. So I think, I think, I think, I think... and I think... and I think... and I think...

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

    summers in Maine ever since.<|endoftext|>Part or all of this entry has been imported from the 19

  2. #2 peak

    of Baden or a fraud? Redeemer of man's sins or "ambulatory automatist"? The curious circumstances and significance of his life have been disputed ever since.⏎In Kaspar, Quebec cartoonist Obom draws on Hauser's own

  3. #3 peak

    Parliament is worth noting. These designs were prepared by him some time in the sixties, and have been in the board of Works ever since. Two or three years ago the suggestion was made that the completion of the panels

  4. #4 peak

    are capable of producing an electromagnetic pulse (EMP) attack if they are nuclear-tipped. This scenario has been a grave concern in the realm of national security ever since the

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. L0:1704 weakly suppresses
  2. L0:1224 weakly suppresses
  3. L0:1301 weakly suppresses
  4. L0:813 weakly excites
  5. L0:178 weakly suppresses
  6. L0:2295 weakly suppresses
  7. L0:2913 weakly suppresses
  8. L0:1147 weakly suppresses
  9. L0:970 weakly suppresses
  10. L0:1313 weakly excites
  11. L0:1580 weakly excites
  12. L0:2272 weakly excites
  13. L0:1657 weakly suppresses
  14. L0:2954 weakly suppresses
  15. L0:1394 weakly suppresses
  16. L0:1382 weakly excites
  17. L0:2949 weakly excites
  18. L0:906 weakly excites 'to' in 'access to'
  19. L0:708 weakly excites
  20. L0:498 weakly suppresses
  21. L0:1894 weakly excites
  22. L0:2243 weakly suppresses
  23. L0:1581 weakly suppresses
  24. L0:2237 weakly excites

Feeds later layers

  1. L17:1282 excites
  2. L4:1787 weakly excites
  3. L3:379 weakly excites 'that' after 'the good news is'
  4. L18:1162 weakly suppresses
  5. L6:61 weakly suppresses
  6. L20:2992 weakly suppresses
  7. L24:704 weakly suppresses
  8. L4:227 weakly excites
  9. L5:2921 weakly suppresses
  10. L14:531 weakly suppresses
  11. L16:1617 weakly suppresses
  12. L2:1374 weakly suppresses
  13. L25:1163 weakly suppresses
  14. L7:751 weakly suppresses
  15. L26:1810 weakly suppresses
  16. L20:698 weakly suppresses
  17. L3:2883 weakly suppresses
  18. L6:2431 weakly excites
  19. L25:1625 weakly suppresses
  20. L7:3014 weakly excites
  21. L23:2165 weakly suppresses
  22. L7:1664 weakly excites
  23. L5:1913 weakly excites
  24. L6:2817 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 -1.31 and -1.191 values between -0.74 and -0.625 values between -0.62 and -0.5114 values between -0.51 and -0.4023 values between -0.40 and -0.28179 values between -0.28 and -0.171005 values between -0.17 and -0.051868 values between -0.05 and 0.06676 values between 0.06 and 0.18107 values between 0.18 and 0.2910 values between 0.29 and 0.403 values between 0.40 and 0.524 values between 0.63 and 0.751 values between 0.75 and 0.861 values between 0.86 and 0.971 values between 0.97 and 1.091 values between 1.20 and 1.322 values between 1.43 and 1.552 values between 1.55 and 1.661 values between 1.89 and 2.001 values between 2.46 and 2.571 values between 2.91 and 3.031 values between 3.14 and 3.26offnegativepositiveoften

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. there
  2. berman
  3. sleeper
  4. asyncio
  5. ooter
  6. CHARSET
  7. apy
  8. -Identifier
  9. setBackground
  10. ikel

Suppresses

  1. figure
  2. gabe
  3. GroupId
  4. gratuita
  5. groupId
  6. pez
  7. ives
  8. figures
  9. ительно
  10. овая