Neuron Explorer

Qwen3-0.6B

blank slot between a date's day and year

A neuron in layer 4 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 spot in a full date just before the year, such as the gap after "May 9," when the year "1862" is about to come next. It is strongest when a full date is written out in news or history text. Turning it up did not give a clear, readable change in the model's output.

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

    security requirements of the PCI Security Standards Council. Outdated browsers do not support the requirements of modern web applications and pose significant security risks. After May 1, 2017, this browser will no longer be supported and you may experience

  2. #2 peak

    across party lines, as this photo was physical evidence of the trauma that children experience during separation. After intense pressure socially and politically, on June 20, 2018, the Trump administration reversed the policy, just eight days after

  3. #3 peak

    the Hutchinson County. The site of Stinnett was selected because it was out of the rough country along the Canadian River.⏎On July 24th, 1926, A. S. Stinnett and Joe Williams paid

  4. #4 peak

    Buffalo, Rochester, Minneapolis, Chicago, and Los Angeles.⏎Eight months after the formation of the Proletarian University of America, on May 24, 1919

  5. #5 peak

    government research money comes from. John Boylan, busy reading medical school applications, replied simply that it comes from NIH. So, on Feb. 9, 1957, Terence sent NIH his grant application:⏎Boylan's

  6. #6 peak

    They battled the elements, disease, hunger and hostile natives.⏎Many died on the journey.⏎But recording the movement of Venus across the sun on June 3, 1769, helped gauge a crucial standard of measurement called the Astronom

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 only way I can you can't you can't you can't you can't you can't you can't you can't you can't you can't you can't you can't you can't you can't you can't

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

    ⏎ENERGY STAR was introduced by the US Environmental Protection Agency in 1992 as a voluntary labeling program designed to identify and promote energy-efficient products, in order to reduce carbon dioxide emissions. EPA partnered with the US Department of Energy in

  2. #2 peak

    medium" ids="2568,2567,2569,2570"]These skills enable them to increase income, in the market, and how to profitably invest their income. We have been working

  3. #3 peak

    Bookworms, and her daughter Havana, we are tackling a controversial topic that we hope will help to

  4. #4 peak

    freedom of scientific research and encourages cooperation in the service of science.⏎Article III: Parties agree to exchange research and program plans, scientific results, and scientific personnel, in the spirit of scientific cooperation.⏎Article IV: Discusses territorial claims and declares that

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. L2:66 excites
  2. L0:2921 suppresses
  3. L3:487 suppresses
  4. L3:1081 excites
  5. L2:704 weakly excites
  6. L1:255 weakly excites "to" after "the best way is"
  7. L1:445 weakly excites document start marker token
  8. L0:1854 weakly excites
  9. L0:2877 weakly suppresses
  10. L3:2117 weakly excites
  11. L3:28 weakly suppresses
  12. L2:1029 weakly excites
  13. L0:2069 weakly excites
  14. L2:482 weakly suppresses
  15. L2:1077 weakly suppresses
  16. L1:1652 weakly excites
  17. L1:385 weakly excites "as" after "known"
  18. L1:2207 weakly excites
  19. L1:1357 weakly excites
  20. L1:2606 weakly suppresses
  21. L3:59 weakly excites
  22. L3:2779 weakly suppresses
  23. L0:610 weakly suppresses
  24. L3:647 weakly suppresses

Feeds later layers

  1. L5:2773 excites
  2. L9:865 excites
  3. L5:259 suppresses
  4. L7:2503 excites
  5. L14:859 excites
  6. L6:2469 suppresses
  7. L7:288 weakly suppresses
  8. L14:155 weakly suppresses sentence-ending periods in prose
  9. L16:1905 weakly suppresses
  10. L7:685 weakly excites
  11. L14:690 weakly suppresses
  12. L5:974 weakly suppresses
  13. L17:386 weakly suppresses
  14. L16:2535 weakly suppresses
  15. L11:58 weakly excites web address colon in urls
  16. L16:1595 weakly excites
  17. L5:1067 weakly suppresses
  18. L24:87 weakly suppresses
  19. L11:2969 weakly excites
  20. L11:2666 weakly suppresses
  21. L5:2280 weakly excites
  22. L18:1085 weakly suppresses full stops ending sentences
  23. L9:288 weakly suppresses
  24. L17:1779 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 -2.90 and -2.761 values between -2.76 and -2.622 values between -2.62 and -2.482 values between -2.48 and -2.331 values between -2.05 and -1.901 values between -1.90 and -1.761 values between -1.33 and -1.194 values between -1.19 and -1.041 values between -1.04 and -0.904 values between -0.90 and -0.764 values between -0.76 and -0.616 values between -0.61 and -0.4713 values between -0.47 and -0.3381 values between -0.33 and -0.19625 values between -0.19 and -0.041948 values between -0.04 and 0.10945 values between 0.10 and 0.24226 values between 0.24 and 0.3933 values between 0.39 and 0.534 values between 0.53 and 0.671 values between 0.67 and 0.821 values between 0.82 and 0.961 values between 1.68 and 1.821 values between 2.39 and 2.531 values between 2.68 and 2.82offnegativepositiveoften

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. bara
  2. 高等
  3. HIP
  4. higher
  5. WC
  6. me
  7. %^
  8. higher
  9. pun
  10. PL

Suppresses

  1. エネルギ
  2. _slave
  3. もらって
  4. ournée
  5. ечно
  6. vous
  7. SelectedItem
  8. .getSelected
  9. @dynamic
  10. -cli