Neuron Explorer

Qwen3-0.6B

'at' in 'at risk'

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

This neuron fires on the word 'at' in the phrase 'at risk', as in 'put their lives at risk'. Turning it up made the model repeat 'at risk' 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

    how to care for people with disabilities, it promotes bias in healthcare. As a result, patients with disabilities lose trust in medical care, which can put their lives at risk.⏎The Impact of Low Accreditation Standards⏎On Dec. 14

  2. #2 peak

    safely handle their vehicles and make prudent travel decisions. When bus drivers do not adhere to proper bus-handling protocol or traffic laws, they put dozens of vulnerable children at risk. The most common causes of fatal school-bus accidents include the following:⏎

  3. #3 peak

    they provide screening services to those communities most at risk for Hepatitis B and C, hypertension and osteoporosis. The collaboration

  4. #4 peak

    - Attempt to reconnect alienated youth or students who are at-risk for behavior problems or violence with the mainstream student body. Use threat-assessment procedures to identify those at risk before the offending behavior is exhibited.⏎

  5. #5 peak

    consists of two different types: open release and endoscopic surgery.⏎(Click Carpal Tunnel Syndrome to learn more, including what causes the condition and who is most at risk for it. You can also click any of the links in the box to

  6. #6 peak

    , cardiogenic shock and death.⏎Increasing evidence of COVID-19’s negative impacts on the cardiovascular system highlights a great need for identifying COVID-19 patients at risk for heart problems, the researchers say. However, no such predictive capabilities currently

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: when it comes to the word "will" in the past perfect willing and now it's used in the past perfect willing, as in the past perfect willing to something, and then it's now used as a verb to have

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

    desert air.⏎So it appears that they have been homophones, at least in the London dialect, for over four hundred years.⏎Wordsworth (1770–1850) also appears to have only used the one-s

  2. #2 peak

    said.⏎The National Atmospheric Release Advisory Center in Livermore also assisted with the 1986 Chernobyl nuclear power plant disaster and the 1979 Three Mile Island accident

  3. #3 peak

    bit here and there, that does them in,” says Jean Lewis, a dietician at St. Clair Hospital.⏎In a study in the New England Journal of Medicine, researchers at Harvard have pinpointed specific foods that contribute to weight gain over

  4. #4 peak

    general manager of Corning Diesel Technologies, the original diesel substrates featured about 1<|endoftext|>The Stevenson House was the childhood home of Robert Louis Stevenson (1850 – 1894), creator of Treasure Island, Kidnapped

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. L21:1439 suppresses
  2. L17:2930 excites
  3. L21:1887 suppresses
  4. L22:1767 suppresses
  5. L21:820 suppresses
  6. L21:1596 suppresses
  7. L20:2160 suppresses
  8. L22:2078 excites
  9. L15:1320 suppresses
  10. L18:2623 suppresses
  11. L20:1678 suppresses
  12. L21:1472 suppresses
  13. L8:7 excites
  14. L12:2115 suppresses
  15. L19:2152 suppresses
  16. L22:1395 excites
  17. L11:1939 suppresses
  18. L4:2398 weakly suppresses
  19. L2:122 weakly excites
  20. L2:75 weakly suppresses
  21. L22:289 weakly excites
  22. L18:569 weakly excites
  23. L11:2800 weakly excites
  24. L3:1531 weakly suppresses

Feeds later layers

  1. L27:599 strongly excites
  2. L26:835 strongly excites
  3. L24:1252 excites
  4. L27:285 excites
  5. L25:2063 excites the word before 'risk'
  6. L24:632 excites
  7. L26:1210 excites
  8. L24:856 excites
  9. L27:835 weakly excites
  10. L27:1258 weakly excites
  11. L25:501 weakly suppresses
  12. L26:840 weakly suppresses
  13. L24:984 weakly excites
  14. L26:941 weakly excites
  15. L25:373 weakly excites
  16. L24:380 weakly suppresses the word 'at' in place phrases
  17. L24:2023 weakly excites
  18. L26:2537 weakly suppresses
  19. L26:2190 weakly suppresses
  20. L24:2202 weakly excites
  21. L25:2595 weakly excites
  22. L24:1485 weakly excites
  23. L26:1950 weakly excites
  24. L25:679 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.

7 values between -9.36 and -5.978 values between -5.97 and -2.583760 values between -2.58 and 0.81100 values between 0.81 and 4.2014 values between 4.20 and 7.5810 values between 7.58 and 10.975 values between 10.97 and 14.361 values between 14.36 and 17.751 values between 17.75 and 21.141 values between 24.53 and 27.921 values between 122.80 and 126.19offnegativepositiveoften

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. risk
  2. 风险
  3. at
  4. risk
  5. Risk
  6. Risk
  7. 風險
  8. risks
  9. 风险管理
  10. -risk

Suppresses

  1. 偃
  2. oub
  3. ota
  4. OTA
  5. otte
  6. trải
  7. 坐下
  8. 棠
  9. allel
  10. ient