Neuron Explorer

Qwen3-0.6B

"press" in "press release"

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 lights up on the word "press" in phrases like "said in a press release," which news stories use all the time. Its opposite side shows up around talk of phones and screens, but that is weaker. Turning it up made the model drift to unrelated topics, so the steering results don't clearly show what it does.

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

    World Bank and the International Energy Agency to suggest that renewables and nuclear “tend to exhibit lock-ins and path dependencies that crowd each other out,” according to a press release from the University of Sussex.⏎Basically, one of the messages of the authors

  2. #2 peak

    so the weather we have in the eastern United States, whether it’s cold or warm, tends to stay with us longer,” said study author Jennifer Francis in a press release.⏎Imp

  3. #3 peak

    gillnets, will help with the management and eventual recovery of these populations,” Jeffrey Mangel, one of the lead authors of the paper, said in a press release.⏎Sea turtles live in Peruvian coastal waters, with multiple species of green

  4. #4 peak

    . West Chicago High School’s Glee Club will hold a “Charity Minstrel Show” on Dec. 21.⏎According to the show’s press releases, “

  5. #5 peak

    some measures, their use was associated with fewer mental health problems.⏎“With these robust findings,” lead author Pål-Ørjan Johansen said in a news release, “it is difficult to see how<|endoftext|>of subpar flows in the

  6. #6 peak

    and treatments, rehabilitation, counseling, a full-time caretaker, inability to drive<|endoftext|>UK hydrogen cars are coming - if you can fill up⏎LONDON (Reuters) - Britain's hydrogen fuel cell car fleet may hit top gear within five years

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 get a few seconds of video and I get a few minutes of news. The problem is that the video and news are mostly the same, and I have to do the task. I get a few seconds of video and I get a

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

    can provide a refreshing break and benefit mental health. Disabling certain apps, using alternative ways of communication, and other tips can help.⏎Many people continually check their phones and other devices for notifications and spend time scrolling through social media. These activities can

  2. #2 peak

    Affecting Our Health?⏎There is plenty of research to suggest that screens are making us less healthy, and it’s easy to see why. Time spent looking at a screen is time spent being sedentary - unless you walk and text,

  3. #3 peak

    we delve into the negative effects of screen time on the quality of your night’s rest. We also share some tips on how to improve your sleep pattern by reducing phone usage before you get into bed.⏎It is universally acknowledged that all adults require seven

  4. #4 peak

    is this. Digital Generation⏎The researchers note that their findings represent the subjective views of teachers and should not be seen as definitive proof that widespread use of computers, phones and video games affects students’ capability to focus.⏎Even so, the researchers who

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. L8:84 suppresses
  2. L9:2723 weakly excites
  3. L9:2295 weakly suppresses
  4. L9:398 weakly suppresses
  5. L8:221 weakly excites
  6. L7:2829 weakly excites
  7. L9:567 weakly excites
  8. L8:2165 weakly suppresses
  9. L6:827 weakly excites
  10. L6:2729 weakly excites
  11. L6:2484 weakly excites
  12. L6:1951 weakly excites
  13. L7:797 weakly excites
  14. L9:2221 weakly excites
  15. L9:585 weakly suppresses
  16. L9:2880 weakly suppresses
  17. L6:1162 weakly suppresses
  18. L7:340 weakly excites
  19. L8:2239 weakly excites
  20. L7:1975 weakly excites
  21. L6:828 weakly suppresses
  22. L7:1155 weakly excites
  23. L2:663 weakly excites
  24. L5:1681 weakly suppresses

Feeds later layers

  1. L19:2886 suppresses
  2. L17:1902 suppresses
  3. L21:1901 suppresses
  4. L20:1426 suppresses
  5. L11:1628 excites
  6. L23:3028 suppresses
  7. L18:2876 weakly suppresses
  8. L20:887 weakly suppresses
  9. L11:1000 weakly suppresses
  10. L21:2659 weakly suppresses
  11. L11:1056 weakly suppresses
  12. L24:537 weakly suppresses
  13. L12:2870 weakly suppresses
  14. L21:1595 weakly excites
  15. L22:1679 weakly suppresses
  16. L20:1132 weakly suppresses
  17. L17:3010 weakly suppresses
  18. L21:2958 weakly excites
  19. L13:2730 weakly suppresses
  20. L11:1498 weakly excites
  21. L26:1315 weakly suppresses
  22. L11:2395 weakly excites
  23. L19:2326 weakly suppresses
  24. L11:452 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.

1 values between -4.31 and -4.151 values between -3.32 and -3.151 values between -3.15 and -2.991 values between -2.99 and -2.822 values between -2.66 and -2.492 values between -2.49 and -2.322 values between -2.32 and -2.161 values between -1.99 and -1.835 values between -1.83 and -1.662 values between -1.49 and -1.335 values between -1.33 and -1.165 values between -1.16 and -1.007 values between -1.00 and -0.838 values between -0.83 and -0.6722 values between -0.67 and -0.5049 values between -0.50 and -0.33296 values between -0.33 and -0.171384 values between -0.17 and -0.001637 values between -0.00 and 0.16403 values between 0.16 and 0.3353 values between 0.33 and 0.508 values between 0.50 and 0.662 values between 0.66 and 0.835 values between 0.83 and 0.992 values between 0.99 and 1.162 values between 1.16 and 1.321 values between 1.32 and 1.491 values between 2.15 and 2.32offnegativepositiveoften

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. 荚
  2. mith
  3. Unters
  4. 彦
  5. tire
  6. stash
  7. Naming
  8. render
  9. difference
  10. صنع

Suppresses

  1. 社交媒体
  2. ertainment
  3. itra
  4. afone
  5. 传媒
  6. uffman
  7. 娱乐
  8. cohol
  9. entertainment
  10. 晒