Neuron Explorer

Qwen3-0.6B

'up' after show, turn and showed

A neuron in layer 2 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 is very clear: it lights up on the word 'up' when it follows verbs like 'show', 'showed' or 'turn', as in 'show up' and 'turn up the heat'. Turning it up did not make the model write about showing or turning anything, so the steering did not reveal the same theme.

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

    -WEFA. “A lot of prominent geologists just laugh at this. There are wolves out there, but if you keep crying wolf and no wolves show up, you start to lose credibility.”⏎The dispute centers on the precise timing of what

  2. #2 peak

    0 times, a torn meniscus will show up on one of<|endoftext|>Under the most lucrative defense contract ever awarded, the Pentagon selected

  3. #3 peak

    meantime, a crowd had begun to gather outside around a woman who had witnessed Rivera's apprehension and was shouting that Rivera was being beaten. When an ambulance showed up to treat the wounds of the employee who had been bitten, it appeared to confirm

  4. #4 peak

    Cyril The Lonely Cloud by Tim Hopgood⏎It’s a bright and lovely day, the perfect day for a picnic, until Cyril the lonely cloud shows up. Everybody agrees he is a bore and a spoilsport and that things are

  5. #5 peak

    as a shock to the couple, who had recently moved from Washington state to their new home in Vegas.⏎A couple of days later, police and investigation officers showed up on their door, as per a local news report. They determined that the bones

  6. #6 peak

    of US foreign policy. Like an American abroad who can’t seem to communicate with the natives when they don’t get the message, his answer is merely to turn up the volume and repeat the message as a tautology: US policy is democratic

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 am a 23-year-old with a history of not knowing. I am currently working toward a medical condition. I am a person with a history of not knowing. I am a person with a history of not knowing. I am

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

    Yellow River Delta World of Change feature.<|endoftext|>Improved Screening for Adverse Childhood Experiences⏎The Institute for Innovation & Implementation at the University of Maryland School of Social Work has partnered with Harvard's Center for the Developing Child's Frontiers of Innovation Community

  2. #2 peak

    34 percent did so after the two-month program.⏎Moises Prospero, a Utah research consultant on criminal and social justice issues with a doctorate in social work, said many students take their messages

  3. #3 peak

    care practitioners surveyed had discussed genetics with their clients. Yet fewer than 10 percent of those people surveyed - including physical therapists, speech and hearing therapists and social workers - were confident in their training in these issues. According to Dr. E.

  4. #4 peak

    help sustain the museum long-term.⏎<|endoftext|>“A 200:1 ratio is very difficult to overcome,” said Darryl Mitteldorf, Licensed Clinical Social Worker (LCSW) and program director of Malecare, a 14

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. L1:2296 weakly suppresses
  2. L1:784 weakly excites
  3. L1:2801 weakly excites
  4. L1:1356 weakly suppresses
  5. L0:2162 weakly suppresses
  6. L1:92 weakly excites
  7. L0:3047 weakly excites
  8. L1:14 weakly suppresses
  9. L1:224 weakly excites
  10. L1:2836 weakly excites
  11. L1:808 weakly suppresses
  12. L0:2960 weakly suppresses
  13. L0:2942 weakly suppresses
  14. L0:60 weakly excites
  15. L1:1770 weakly suppresses
  16. L1:2176 weakly excites
  17. L1:2116 weakly suppresses
  18. L1:2234 weakly suppresses
  19. L0:818 weakly suppresses
  20. L1:1984 weakly suppresses
  21. L1:1686 weakly suppresses
  22. L1:134 weakly excites
  23. L1:87 weakly suppresses
  24. L1:1464 weakly excites

Feeds later layers

  1. L7:2568 suppresses
  2. L13:1871 weakly suppresses
  3. L10:2699 weakly suppresses
  4. L25:2602 weakly suppresses
  5. L8:2418 weakly suppresses
  6. L13:2082 weakly suppresses
  7. L6:3058 weakly suppresses
  8. L11:2731 weakly suppresses the word 'course' in course blurbs
  9. L11:2480 weakly excites
  10. L5:1584 weakly suppresses
  11. L22:1949 weakly suppresses
  12. L7:1823 weakly suppresses
  13. L5:2781 weakly suppresses
  14. L13:1540 weakly suppresses
  15. L21:2178 weakly suppresses
  16. L6:1559 weakly suppresses
  17. L21:2125 weakly suppresses
  18. L11:1139 weakly excites
  19. L11:801 weakly suppresses
  20. L14:2681 weakly suppresses
  21. L20:1500 weakly suppresses
  22. L8:2476 weakly suppresses
  23. L23:2081 weakly suppresses
  24. L16:2688 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.

2 values between -0.66 and -0.581 values between -0.58 and -0.494 values between -0.49 and -0.4114 values between -0.41 and -0.3221 values between -0.32 and -0.24143 values between -0.24 and -0.15478 values between -0.15 and -0.071588 values between -0.07 and 0.011165 values between 0.01 and 0.10361 values between 0.10 and 0.1891 values between 0.18 and 0.2729 values between 0.27 and 0.352 values between 0.35 and 0.443 values between 0.44 and 0.521 values between 0.52 and 0.601 values between 0.60 and 0.691 values between 0.69 and 0.772 values between 0.77 and 0.861 values between 2.62 and 2.71offnegativepositiveoften

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. icles
  2. heet
  3. bosses
  4. افة
  5. acles
  6. ورية
  7. -shirt
  8. ้ง
  9. .comments
  10. above

Suppresses

  1. 眼角
  2. edException
  3. ,$
  4. }⏎⏎⏎⏎⏎
  5. antages
  6. misdemean
  7. //}⏎⏎
  8. ;\⏎
  9. hipster
  10. 篱