Neuron Explorer

Qwen3-0.6B

'had' or 'has' after a subject (perfect tense)

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 fires on the word 'had' in phrases like 'I've had' or 'Teachers have had to', which is the past participle used in perfect tenses. It seems to switch off on 'has' in some sentences, but the pattern is only moderately clear. Turning it up made the model's text drift toward unrelated topics.

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

    can be utilized, with rye and millet being probably the most well-known. Different decisions are corn, wheat, and various cereal grains. I’ve even had the best of luck with popcorn!⏎Grain can make sawdust generate more

  2. #2 peak

    of empowering individuals in these everyday domains are much more likely to maintain a vibrant democracy in traditional political settings.⏎I might blog about it next week, after I’ve had a chance to read it, but one finding caught my eye.⏎People across the

  3. #3 peak

    respond to them very well. With their education in mind, schools and nurseries all over the world have resorted to online learning. Teachers have had to rework their teaching methods and try to engage children through the screens. They

  4. #4 peak

    investigate the CIA and to make sure that they are not engaging in activities that are not authorized by law.⏎Who has oversight over the CIA⏎The US Congress has had oversight responsibilities over the CIA since the Agency was established in 1947

  5. #5 peak

    . She struggles with some behavioral problems at the home and has difficulty communicating her needs to the staff.⏎Dr. Smith remarks, “I notice that Stephanie has never had a pap smear before, at least according to our records. I wanted to check

  6. #6 peak

    have found disturbing concentrations of pesticides in some baby foods. And the U.S. Department of Agriculture has found weed killers widespread in finished tap water. Environmentalists have had important successes in forcing pesticides that presented unacceptably high dietary risks off the market

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: 12, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1

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

    Published: 23rd Dec 2011 08:41:37⏎The prospect of untreatable gonorrhoea has provoked alarm around the world, and there are no new classes of antibiotics in development

  2. #2 peak

    , an entomologist at the Smithsonian Tropical Research Institute and father of two of the kids, had been tending to this tree, and was worried that it had suffered irreparable damage. But when he returned later, he was in for a

  3. #3 peak

    subspecies (type) of milk snake is a method of helping control body temperature.<|endoftext|>ABC News reports:⏎Public health officials have announced that a squirrel in Colorado has tested positive for the bubonic plague. The town of Morrison, Colorado, in

  4. #4 peak

    5 inches. Slopes range from 4 to 60 percent.<|endoftext|>RIVERSIDE, Calif. – Biologists at the University of California, Riverside have found that voluntary activity, such as daily exercise, is a highly heritable trait

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. L3:2249 weakly excites
  2. L1:2434 weakly excites
  3. L3:115 weakly suppresses
  4. L1:1344 weakly excites 'who' after a person noun
  5. L3:186 weakly suppresses
  6. L2:2353 weakly excites
  7. L3:691 weakly excites
  8. L2:237 weakly excites
  9. L0:1701 weakly suppresses
  10. L3:2438 weakly excites
  11. L3:2701 weakly excites
  12. L3:1128 weakly suppresses
  13. L0:2258 weakly excites
  14. L3:2318 weakly suppresses
  15. L3:2270 weakly excites
  16. L2:653 weakly suppresses
  17. L3:748 weakly suppresses
  18. L1:2054 weakly excites
  19. L0:376 weakly excites
  20. L2:2426 weakly suppresses
  21. L1:1464 weakly excites
  22. L0:890 weakly excites
  23. L3:460 weakly excites
  24. L3:2082 weakly excites

Feeds later layers

  1. L11:1644 suppresses
  2. L11:1669 excites
  3. L8:857 weakly suppresses
  4. L19:1684 weakly suppresses
  5. L7:1243 weakly suppresses
  6. L7:654 weakly suppresses
  7. L12:2574 weakly excites
  8. L11:2433 weakly suppresses
  9. L5:237 weakly excites
  10. L5:717 weakly excites
  11. L12:1671 weakly suppresses
  12. L7:681 weakly suppresses 'from' after prevent or stop
  13. L14:1908 weakly suppresses
  14. L10:2652 weakly suppresses
  15. L15:2320 weakly suppresses
  16. L15:2435 weakly suppresses unclear: weak mid-sentence function words
  17. L22:2864 weakly excites
  18. L23:2852 weakly excites
  19. L5:1909 weakly suppresses
  20. L9:2088 weakly excites
  21. L23:1447 weakly excites
  22. L17:2285 weakly suppresses
  23. L16:2678 weakly excites
  24. L6:1821 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 -6.42 and -6.131 values between -3.83 and -3.541 values between -3.54 and -3.252 values between -2.97 and -2.681 values between -2.68 and -2.393 values between -2.39 and -2.102 values between -2.10 and -1.8111 values between -1.81 and -1.536 values between -1.53 and -1.249 values between -1.24 and -0.9520 values between -0.95 and -0.6635 values between -0.66 and -0.37748 values between -0.37 and -0.092864 values between -0.09 and 0.20161 values between 0.20 and 0.4918 values between 0.49 and 0.786 values between 0.78 and 1.079 values between 1.07 and 1.353 values between 1.35 and 1.642 values between 1.93 and 2.223 values between 2.22 and 2.511 values between 2.79 and 3.081 values between 4.81 and 5.10offnegativepositiveoften

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. ereal
  2. ampil
  3. esson
  4. 锋
  5. itary
  6. edith
  7. later
  8. .gc
  9. coords
  10. snap

Suppresses

  1. icient
  2. iability
  3. 请您
  4. .Dimension
  5. utherland
  6. _attr
  7. ptive
  8. ombs
  9. *⏎⏎
  10. impairment