Neuron Explorer

Qwen3-0.6B

comma before a trailing 'with' clause

A neuron in layer 23 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

The neuron fires on a comma that ends a full clause in news-style writing, where the sentence is about to continue with a phrase introduced by 'with'. When turned up, the model just repeated the word 'with' 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

    wave in northern Europe was made twice as likely by climate change, according to a preliminary analysis.⏎Temperatures have soared over much of Europe over the last month, regularly exceeding 30

  2. #2 peak

    than occupants of cars, and five times more likely to be injured.⏎Helmet usage is also on the decline, dropping from 66 percent of motorcyclists wearing helmets in 20

  3. #3 peak

    costs businesses approximately $7 billion per year due to the nearly 111 million work days lost. Norovirus cases are also on the rise this season, and the CDC reports that foodborne norovirus illness accounts for

  4. #4 peak

    .⏎The first step in winning the lottery is to purchase a ticket, which can be done online or at any store that sells state-approved tickets. Ticket prices vary, but most are inexpensive. Some are scratch-off cards, while others are more expensive

  5. #5 peak

    infections.⏎Sub-Saharan Africa remains most severely affected, with nearly 1 in every 25 adults living with HIV and accounting for

  6. #6 peak

    ish meat, and fish. Participants were ranked by DASH score into five groups, or quintiles.⏎The researchers found that higher DASH scores were associated with higher

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: if the main part of the world is a web of interconnected systems, such as the internet, then the main elements are all interconnected, and the main elements are not just one particular system. So if the main part is a web of systems

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

    our Patrons Cristina Magistrali, Toren Wallengren, Eric Huffman, Julia Casey, Colton Siefker, Daniel Wenger, Matias Mancini, John Thompson, Chris Krish, and Alvero Wiggins for supporting

  2. #2 peak

    the second method is employed for large a.c. generators⏎!! OOPS Login [Click here] is required for more results / answer<|endoftext|>By: Swapnil Katiyar⏎With the 17th Lok Sabha elections round the corner,

  3. #3 peak

    much” is the underlying philosophy of a number of innovative programmes being carried out in Uruguayan schools that are using gardens as a teaching resource, explained Edith Moraes, director of the national Primary Education Board.⏎One of these initiatives is

  4. #4 peak

    x 66 mm|⏎|Printer||American Bank Note Company (U.S.A.) – ABNC|⏎|Signatures||Ernesto (Che) Guevara Serna & Rolando Díaz Astaraín|⏎|Comment

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. L19:1036 suppresses
  2. L20:1832 suppresses 'as' introducing a reason
  3. L19:1714 suppresses
  4. L18:1530 suppresses
  5. L20:476 excites commas that lead into a continuation
  6. L22:992 weakly suppresses
  7. L18:1514 weakly suppresses
  8. L6:2407 weakly suppresses
  9. L20:1043 weakly excites
  10. L22:1489 weakly excites
  11. L14:1634 weakly excites
  12. L13:2296 weakly suppresses
  13. L19:612 weakly excites
  14. L22:2630 weakly excites
  15. L9:1541 weakly excites
  16. L2:215 weakly suppresses
  17. L18:753 weakly suppresses
  18. L7:1471 weakly suppresses
  19. L10:1638 weakly suppresses
  20. L18:366 weakly excites comma inside numbers and dates
  21. L11:1047 weakly excites pieces of software and library names
  22. L17:1537 weakly excites
  23. L16:846 weakly excites comma opening an aside or explanation
  24. L8:1366 weakly excites

Feeds later layers

  1. L24:921 strongly excites
  2. L26:741 strongly excites
  3. L27:603 strongly excites
  4. L24:883 strongly excites
  5. L27:1443 excites
  6. L26:223 suppresses
  7. L27:2153 excites
  8. L25:2139 suppresses capital N starting an abbreviation
  9. L24:2419 weakly suppresses
  10. L24:1002 weakly excites
  11. L25:850 weakly suppresses
  12. L25:2803 weakly excites
  13. L24:394 weakly suppresses
  14. L24:360 weakly excites
  15. L25:1564 weakly excites word pieces before 'trans' or 'transfer'
  16. L25:230 weakly excites
  17. L27:2862 weakly suppresses
  18. L24:1593 weakly excites
  19. L27:1869 weakly suppresses
  20. L27:1815 weakly suppresses
  21. L24:670 weakly suppresses
  22. L24:1666 weakly excites
  23. L26:2062 weakly suppresses
  24. L26:973 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.

2 values between -5.39 and -4.591 values between -4.59 and -3.787 values between -3.78 and -2.976 values between -2.97 and -2.1712 values between -2.17 and -1.36179 values between -1.36 and -0.563078 values between -0.56 and 0.25537 values between 0.25 and 1.0534 values between 1.05 and 1.8617 values between 1.86 and 2.667 values between 2.66 and 3.475 values between 3.47 and 4.285 values between 4.28 and 5.083 values between 5.08 and 5.893 values between 5.89 and 6.696 values between 6.69 and 7.501 values between 7.50 and 8.301 values between 8.30 and 9.111 values between 10.72 and 11.521 values between 11.52 and 12.331 values between 14.75 and 15.551 values between 26.02 and 26.83offnegativepositiveoften

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. with
  2. with
  3. With
  4. With
  5. _with
  6. →with
  7. .with
  8. avec
  9. WITH
  10. -with

Suppresses

  1. haled
  2. edar
  3. ewing
  4. ода
  5. 妫
  6. ül
  7. .createFrom
  8. urgeon
  9. iedo
  10. 并对