Neuron Explorer

Qwen3-0.6B

word stems before a suffix, 'dead' negative

A neuron in layer 0 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron tends to fire on the first part of a word that is followed by an ending, like 'MSD' in 'MSDS' or 'disput' in 'disputing'. It is strongly pushed down by the word 'dead', as in 'dead elephant' or 'dead ringer'. The turned up text did not show a clear related theme.

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

    communicate information on these hazards. The information contained in the SDS is largely the same as the MSDS, except now the SDSs are required to be presented in a consistent user

  2. #2 peak

    0(g)), revised in 2012, requires that the chemical manufacturer, distributor, or importer provide Safety Data Sheets (SDSs) (formerly MSDSs or Material Safety Data Sheets) for each hazardous chemical to downstream users to

  3. #3 peak

    0, 2010 - Connecticut teachers visited Quinnipiac University during the summer of 2010 to brush up on classroom technology. The Bristol-Myers Squibb Center for Science Teaching and Learning at Quinnipiac

  4. #4 peak

    threatened to execute or sell them into slavery. This broadside reassured potential black recruits that the U.S. Government would treat all of its troops as soldiers and retaliate in the event of Confederate mistreatment of black U

  5. #5 peak

    process known as alternative dispute resolution (ADR). Mediators, also known as arbitrators and conciliators, guide the ADR process and help resolve conflicts between disputing parties.⏎Job Duties of Mediators:⏎Mediators facilitate negotiation and settlement

  6. #6 peak

    14, 2007 - Scientists Elucidate the Origin of the Darkest Galaxies in the Universe - Press Release⏎-- return to Press Releases --⏎Date Issued: February 14, 2007⏎

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: to make a new life. The only way, and the only way. So I think I think, I think. So I do I do, I do. So I live, I live, I live. So I think 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

    to reduce a⏎dead elephant to nothing more than a pile of bones (see video).⏎remarkable footage

  2. #2 peak

    coffin with an emergency alarm, intercom, a light, breathing apparatus and a heart monitor in 1995.⏎While urban legend says phrases like “dead ringer”, “saved by the bell” and “graveyard shift” are

  3. #3 peak

    locust lumber are not common. A wood boring insect often invades the tree. A leaf miner often turns entire hillsides into brown-leafed, “dead-looking” trees for a week or so in the summer, but this damage is

  4. #4 peak

    80% nitrogen, 16% oxygen, and⏎the rest is a mixture of other various gases.<|endoftext|>WASHINGTON — Scientists predicting the biggest-ever “dead zone” in the Gulf of Mexico are blaming Midwestern corn grown for ethanol for

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

None: this is the first layer.

Feeds later layers

  1. L25:2739 suppresses
  2. L16:269 weakly suppresses
  3. L10:40 weakly suppresses
  4. L10:451 weakly suppresses
  5. L3:2451 weakly suppresses
  6. L23:2276 weakly suppresses
  7. L19:2966 weakly suppresses
  8. L7:1895 weakly excites
  9. L23:3006 weakly suppresses
  10. L24:2146 weakly suppresses
  11. L16:1116 weakly suppresses
  12. L18:372 weakly suppresses
  13. L26:798 weakly suppresses
  14. L23:2869 weakly suppresses
  15. L25:1459 weakly suppresses
  16. L24:40 weakly excites
  17. L13:1376 weakly suppresses
  18. L11:2823 weakly suppresses
  19. L23:903 weakly suppresses
  20. L21:916 weakly suppresses
  21. L13:2764 weakly suppresses
  22. L7:2468 weakly suppresses 'real' as opposed to mock or simulated
  23. L10:1902 weakly excites
  24. L15:2760 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 -1.93 and -1.861 values between -1.00 and -0.931 values between -0.87 and -0.801 values between -0.80 and -0.732 values between -0.73 and -0.674 values between -0.67 and -0.601 values between -0.60 and -0.541 values between -0.54 and -0.473 values between -0.47 and -0.403 values between -0.40 and -0.344 values between -0.34 and -0.2726 values between -0.27 and -0.2160 values between -0.21 and -0.14210 values between -0.14 and -0.071162 values between -0.07 and -0.011931 values between -0.01 and 0.06362 values between 0.06 and 0.13100 values between 0.13 and 0.1917 values between 0.19 and 0.266 values between 0.26 and 0.327 values between 0.32 and 0.392 values between 0.46 and 0.521 values between 0.52 and 0.591 values between 0.59 and 0.661 values between 0.66 and 0.72offnegativepositiveoften

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. чем
  2. окумент
  3. �单
  4. llum
  5. trif
  6. Aug
  7. term
  8. TouchEvent
  9. createView
  10. tro

Suppresses

  1. ündig
  2. houses
  3. fsp
  4. anst
  5. Disposition
  6. enames
  7. wil
  8. hint
  9. ственный
  10. ively