Neuron Explorer

Qwen3-0.6B

opening paren before an acronym

A neuron in layer 20 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 open bracket that comes right after a long organisation or program name, just before its short form, such as 'European Commission (EAPA)'. Turning it up made the model's text wander into abstract, unrelated sentences.

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

    of three concepts undergoing preliminary studies for the three-year, €7 million ($9 million) European hypersonic research project, Long Term Advanced Propulsion Concepts and Technologies (LAPCAT).⏎With €4 million from the European Commission and the rest from

  2. #2 peak

    point the way towards evidence-based solutions that are effective, scalable, and invite implementation.”⏎Alongside Congress and the CFPB, the Native American Financial Services Association (NAFSA) is dedicated to providing financial literacy to everyone. Through their Financial

  3. #3 peak

    as the National Qualification for Headship (NPQH) in the United Kingdom and the (Interstate School Leaders Licensure Consortium (ISLLC) standards in the United States. Distributed patterns of leadership have developed

  4. #4 peak

    Street, a home where teen boys learn independent living skills; a community-based Diagnostic Evaluation service that helps people of all ages; SouthEast Treatment and Reentry Team (START), a program that works with teen girls in residential placement through their transition to

  5. #5 peak

    In early December, the East African Paediatric Association (EAPA) and Kenya Paediatric Assocation (KPA) in collaboration

  6. #6 peak

    sediment core incubations.⏎26 Related JoVE Articles!⏎Assessment of Social Cognition in Non-human Primates Using a Network of Computerized Automated Learning Device (ALDM) Test Systems⏎Institutions: Aix-Marseille University.⏎F

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: the fact that a frun in a frun is a frun in a frun. This is the frun in the frun. This is a frun in a frun. And this is a frun in a fr

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

    Unlimited right to asylum was conceived in West Germany immediately after the Second World War. By adopting this permissive approach to the right to asylum, the Parliamentary Council (Parlamentarischer Rat) wanted to make a deliberate break with the National Socialist

  2. #2 peak

    several financial institutions in conjunction with the Bank of Canada.⏎association canadienne des paiements⏎Related Terms and Acronyms⏎- Clearing and Settlement (C&S) — Acronym,⏎- The process whereby banks collect or pay out for

  3. #3 peak

    Permanent Structured Cooperation (PESCO), which aims to improve European defense capabilities and cooperation with an emphasis on joint defense-related research projects; The European Defence Fund (EDF), a mechanism for funding PESCO projects and incentivizing greater cohesion and efficiency

  4. #4 peak

    potentially dangerous microorganisms. Through a HEPA filter, the continuous airflow is released into the atmosphere. Biological Safety Cabinets, sometimes referred to as Biosafety Cabinets (BSC) or Microbiological Safety Cabinets, are vented laboratory workspaces 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

  1. L19:1829 strongly suppresses opening bracket before an acronym
  2. L17:2755 suppresses
  3. L17:2502 suppresses opening parenthesis before an abbreviation
  4. L17:1307 excites parenthesis introducing an acronym
  5. L14:2419 excites
  6. L19:1778 excites
  7. L19:1011 excites
  8. L17:2698 suppresses
  9. L17:2032 suppresses
  10. L19:2045 excites abbreviation pieces in organisation names
  11. L7:765 suppresses
  12. L18:2674 suppresses
  13. L19:2278 excites
  14. L19:1520 excites
  15. L9:2946 suppresses
  16. L19:2801 suppresses
  17. L11:651 suppresses
  18. L16:722 weakly suppresses
  19. L13:3032 weakly excites
  20. L18:1031 weakly excites
  21. L12:1340 weakly excites
  22. L8:2045 weakly suppresses
  23. L17:1404 weakly suppresses
  24. L12:2126 weakly suppresses

Feeds later layers

  1. L21:496 strongly excites
  2. L21:1291 suppresses
  3. L25:652 suppresses
  4. L23:957 suppresses
  5. L24:555 excites
  6. L25:2204 suppresses
  7. L23:672 suppresses
  8. L25:735 excites
  9. L24:142 suppresses
  10. L25:1105 suppresses
  11. L25:609 suppresses
  12. L26:443 suppresses
  13. L23:2426 suppresses
  14. L26:914 suppresses
  15. L24:1278 excites
  16. L27:129 suppresses
  17. L23:404 excites
  18. L21:1355 weakly excites
  19. L27:266 weakly excites
  20. L23:225 weakly suppresses
  21. L26:23 weakly excites
  22. L26:386 weakly suppresses
  23. L21:857 weakly excites
  24. L25:486 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 -17.28 and -16.211 values between -15.14 and -14.071 values between -13.00 and -11.931 values between -7.66 and -6.591 values between -6.59 and -5.522 values between -5.52 and -4.452 values between -4.45 and -3.383 values between -3.38 and -2.318 values between -2.31 and -1.24784 values between -1.24 and -0.173087 values between -0.17 and 0.9011 values between 0.90 and 1.971 values between 1.97 and 3.041 values between 3.04 and 4.111 values between 4.11 and 5.181 values between 12.67 and 13.741 values between 16.94 and 18.011 values between 24.43 and 25.50offnegativepositiveoften

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. edException
  2. AAC
  3. TEE
  4. theless
  5. LGPL
  6. MCC
  7. =DB
  8. BDS
  9. DWC
  10. -DD

Suppresses

  1. ÜN
  2. mod
  3. too
  4. хо
  5. ove
  6. 爷
  7. avail
  8. com
  9. 大的
  10. tran