Neuron Explorer

Qwen3-0.6B

the '6' in years like 196x

A neuron in layer 2 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 digit 6 when it is the third digit of a year in the 1960s, such as 1966 or 1963. It is a narrow pattern for a specific digit in dates. Turning it up did not change the written text in a way that matches this.

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

    Israel?⏎One part of the wider Black Hebrew Israelite movement began arriving in Israel in the late 1960s, where they settled in the Negev Desert town of Dimona. There were tensions at first between Israeli authorities and

  2. #2 peak

    943, 1949, 1955, 1959 and 1967.⏎ACT Government: Multi-layered online mapping, imagery and land information for

  3. #3 peak

    of Dallas Medical Department.⏎1943-Moved to Houston.⏎1969-Given independent status.⏎1903-College of Pharmacy organized in

  4. #4 peak

    the 1950s and 1960s, and the Free World’s desire to be more well-informed about

  5. #5 peak

    will [reveal them] and they will lose their job,” she said.⏎Baby boomers, or those born between 1946 and 1964, are more vulnerable to the Hepatitis C virus.⏎No vaccine is available

  6. #6 peak

    was a founder of Liberty Bank, the first black-owned bank in Seattle and was the first black president of the Church Council of Greater Seattle.⏎In 1966 McKinney founded and served as president and CEO of the Seattle Opportunity Industrialization

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: I have been thinking about this. I have been thinking about this. I have been thinking about this. I have been thinking about this. I have been thinking about this. I have been thinking about this. I have been thinking about this

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

    latest USDA tests show we have much more work to do.”⏎EWG researchers analyzed annual pesticide residue tests conducted by the USDA and federal Food and Drug Administration

  2. #2 peak

    eighth edition of its Shopper’s Guide to Pesticides in Produce with updated information on 45 popular fruits and vegetables and their total pesticide loads. EWG highlights the worst offenders with its new Dirty Dozen Plus list and the cleanest

  3. #3 peak

    don’t like to eat food contaminated by pesticides,” said EWG president Ken Cook. “Our shopper’s guide to pesticides in produce gives consumers easy

  4. #4 peak

    1957, Terence sent NIH his grant application:⏎Boylan's letter made its way to Ernest M. Allen, head of NIH's fledgling Division of Research Grants and one of the key architects of the agency's peer

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. L1:24 suppresses
  2. L1:2932 weakly excites
  3. L0:54 weakly excites
  4. L1:141 weakly suppresses
  5. L0:2471 weakly excites
  6. L0:770 weakly suppresses "of" after minority, tells or news
  7. L0:2617 weakly suppresses
  8. L1:503 weakly suppresses
  9. L0:594 weakly excites
  10. L0:2775 weakly suppresses
  11. L1:11 weakly excites
  12. L1:1228 weakly excites
  13. L0:782 weakly excites
  14. L0:1010 weakly suppresses
  15. L0:2404 weakly suppresses
  16. L1:1790 weakly suppresses
  17. L0:997 weakly suppresses
  18. L0:858 weakly suppresses
  19. L0:1887 weakly suppresses
  20. L1:1206 weakly excites
  21. L1:1931 weakly excites
  22. L0:1108 weakly excites
  23. L0:1719 weakly suppresses
  24. L0:2977 weakly suppresses

Feeds later layers

  1. L17:2621 suppresses
  2. L3:2595 suppresses
  3. L15:32 weakly excites
  4. L6:2852 weakly suppresses
  5. L10:1094 weakly suppresses
  6. L21:2531 weakly excites
  7. L12:560 weakly excites
  8. L3:2525 weakly suppresses
  9. L7:683 weakly suppresses
  10. L8:2729 weakly suppresses
  11. L24:1728 weakly excites
  12. L12:325 weakly excites
  13. L5:2698 weakly suppresses
  14. L11:2411 weakly suppresses
  15. L19:19 weakly suppresses unclear: words mid-phrase before a continuation
  16. L11:1758 weakly suppresses
  17. L9:1040 weakly excites
  18. L18:404 weakly excites 'get', 'go' and 'got' after a subject
  19. L12:395 weakly excites
  20. L15:1098 weakly excites
  21. L20:2020 weakly suppresses comma before a participle phrase
  22. L16:1877 weakly suppresses
  23. L18:2764 weakly suppresses
  24. L5:2684 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 -1.60 and -1.501 values between -1.00 and -0.911 values between -0.81 and -0.712 values between -0.71 and -0.613 values between -0.61 and -0.5111 values between -0.51 and -0.4113 values between -0.41 and -0.3183 values between -0.31 and -0.21326 values between -0.21 and -0.111095 values between -0.11 and -0.011705 values between -0.01 and 0.09512 values between 0.09 and 0.19120 values between 0.19 and 0.2913 values between 0.29 and 0.399 values between 0.39 and 0.492 values between 0.49 and 0.593 values between 0.59 and 0.693 values between 0.69 and 0.791 values between 0.89 and 0.992 values between 1.09 and 1.191 values between 1.19 and 1.291 values between 2.29 and 2.39offnegativepositiveoften

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. ros
  2. oker
  3. роме
  4. 工业企业
  5. 马拉
  6. ever
  7. 씀
  8. nowled
  9. rotein
  10. chorus

Suppresses

  1. evacuate
  2. bolt
  3. gard
  4. ilater
  5. ese
  6. 鮮
  7. GetValue
  8. return
  9. vale
  10. _customize