Neuron Explorer

Qwen3-0.6B

opening parenthesis before an acronym

A neuron in layer 27 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 fires on the opening bracket that comes right after a full name, just before its short form like (ISM) or (ARS). Turning it down makes the model repeat short capitalised words like "SAM" and "RAM".

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

    have their influence on one another visually mapped. This methodology was named Interpretive Structural Modelling (ISM). Over the next 20 years, the use of ISM to

  2. #2 peak

    Technologies.⏎Dr Copeland and co-investigator Dr Carl Sangan will assess the technical feasibility of producing high temperature radial turbines using the Selective Laser Melting (SLM) Additive Manufacturing (AM) technique with nickel super-alloys.⏎

  3. #3 peak

    So, for a language to be type safe, we have to provide another alternative to the CTS, which is the Common Language Specification.⏎- Common Language Specifications (CLS):⏎CLS is a set of rules that describes, in detail, the minimal

  4. #4 peak

    for suicidal thoughts.⏎Most treatments for suicidal thoughts, including anti-depressants, talk therapy, electroconvulsive therapy (ECT), and transcranial magnetic stimulation (TMS), take several weeks

  5. #5 peak

    , and living or life skills training.⏎- Eligible California residents may connect with the San Andreas Regional Center (SARC) to create an Individualized Program Plan (IPP). This is a contract between SARC and the individual that sets up an

  6. #6 peak

    patriarch towards the defence of the Holy Land. The struggle ended in the deposition of Dagobert in 1102.⏎<|endoftext|>Agricultural Research Service (ARS) scientists have found environmentally and economically sound uses for

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: SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM SAM

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 what kids today consider a major issue. It’s vital that parents understand cyberbullying and find ways to protect children from it.⏎WHAT IS CYBERBULLYING?⏎Cyberbullying takes a variety of forms, but it

  2. #2 peak

    creation and production of theatre.⏎Table of Contents⏎- PRODUCTION AND BACKSTAGE ORGANIZATION⏎- PRODUCING THEATER⏎- CREATING THEATER⏎- THE CREATIVE TEAM⏎- THE CREATIVE ARTISTS⏎

  3. #3 peak

    minute, allowing a shooter to kill dozens of people within a matter of seconds. See George C. Wilson, Visible Violence, 12 NAT’L J. 886, 887 (2003). Short

  4. #4 peak

    , WIND, (shelter?) Freedom from: weeds, pests, disease.⏎- WATER: tap, butts, drainage (water well weekly)⏎- SOIL type, texture, structure, pH, organic matter content, fertility;⏎-

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. L24:292 strongly excites
  2. L25:157 strongly excites
  3. L24:638 strongly suppresses
  4. L26:23 strongly suppresses
  5. L20:366 strongly suppresses
  6. L19:1340 strongly excites
  7. L22:776 strongly suppresses
  8. L21:533 excites
  9. L24:620 excites
  10. L18:599 suppresses
  11. L22:2753 suppresses
  12. L16:584 excites
  13. L24:555 excites
  14. L14:479 suppresses
  15. L25:735 suppresses
  16. L24:749 suppresses
  17. L26:1119 suppresses
  18. L21:753 excites
  19. L25:616 excites
  20. L20:351 suppresses
  21. L18:2566 excites
  22. L24:1991 excites
  23. L10:942 excites
  24. L24:1278 weakly suppresses

Feeds later layers

None: this is the last layer.

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 -24.78 and -23.521 values between -23.52 and -22.261 values between -22.26 and -21.002 values between -18.47 and -17.211 values between -17.21 and -15.951 values between -14.68 and -13.422 values between -13.42 and -12.162 values between -10.90 and -9.643 values between -9.64 and -8.372 values between -8.37 and -7.114 values between -5.85 and -4.598 values between -4.59 and -3.338 values between -3.33 and -2.06105 values between -2.06 and -0.803484 values between -0.80 and 0.46264 values between 0.46 and 1.726 values between 1.72 and 2.994 values between 2.99 and 4.251 values between 4.25 and 5.512 values between 5.51 and 6.772 values between 6.77 and 8.031 values between 8.03 and 9.301 values between 9.30 and 10.561 values between 18.13 and 19.391 values between 24.44 and 25.70offnegativepositiveoften

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. -Re
  2. 引っ越し
  3. 觎
  4. jk
  5. ווה
  6. Jr
  7. ssue
  8. yster
  9. feito
  10. 伟大复兴

Suppresses

  1. SAM
  2. AAC
  3. CAM
  4. MAC
  5. SAT
  6. CAP
  7. RAM
  8. SAC
  9. AMP
  10. SPA