Neuron Explorer

Qwen3-0.6B

'to' before 'be' after considered

A neuron in layer 18 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 'to' when it comes right before 'be', as in 'is considered to be' or 'determined to be'. Turning it up made the model drift toward words like 'being' and 'having'.

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

    8: Tier 2—Assessment and Problem Solving⏎Tier 2—Assessment and Problem Solving⏎We can no longer implement instruction we believe to be great and cross our fingers and hope for the best. We must identify appropriate

  2. #2 peak

    .⏎The constraints and the objective are linear functions.⏎Only problems that satisfy the above requirements can be solved by using the linear programming models. Once linear programming is found to be applicable to the problem, the next step is to formulate the problem.⏎Email

  3. #3 peak

    the ability to trigger programmed cell death and inhibit the action of protein molecules that are responsible for cancer.⏎5} Stimulates the digestive system⏎Ginger is considered to be a good remedy for digestion related disorders.<|endoftext|>Measles is a highly

  4. #4 peak

    with a blood alcohol concentration (BAC) under .08 percent is considered a legal activity in all 50 states. In fact, it’s considered to be more dangerous to drive while talking on a hands-free cell phone than it is

  5. #5 peak

    show.<|endoftext|>The Scarlet Letter is an 1850 romantic work of fiction in a historical setting, written by Nathaniel Hawthorne, and is considered to be his magnum opus. Set in 17th-century Puritan

  6. #6 peak

    to be the science of healing. Ayurveda takes it that the web of life is intricately interwoven and interconnected. Ayurvedic medicine is considered to be one of the world’s oldest medical systems and customizes preventive wellness to the

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 result of a climatology group's report, which claims to say as clearly as possible, as a whole, which result claims to call the result of a climatology group’s report, which claims to as clearly as possible, which result

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

    increase the risk of fractures, bleeds and other serious injuries, so it’s important to tell your doctor if you’re having this symptom. Your doctor can determine the cause and get you appropriate treatment.⏎Here are some of the medical conditions that can

  2. #2 peak

    the public. If you find prior art that shows any of this method is not new or is obvious to a person trained in the art, then let the world know.<|endoftext|>Should you breakfast like a king, lunch like a prince, and dine

  3. #3 peak

    or your child has experienced an allergic reaction to foods, call our office to schedule an appointment with one of our allergy experts so we may perform allergy testing to determine the source of the reaction.⏎For more information visit any of the following sites:<|endoftext|>

  4. #4 peak

    their hats when she passed.⏎Deaconess Alexander wrote, “I am to see everyone gets what they need…some folk don’t need help now and I know who they are. The old people and the children, they need the most…

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. L14:2460 suppresses
  2. L11:1774 excites
  3. L16:674 excites
  4. L17:2426 suppresses 'that' after nouns of belief or fact
  5. L17:2582 excites
  6. L16:2097 suppresses
  7. L6:2450 suppresses 'guess' and 'obvious' after 'as you can see'
  8. L5:3034 suppresses
  9. L3:1805 weakly excites "to" after "found" or "said"
  10. L13:1164 weakly suppresses
  11. L16:1482 weakly excites
  12. L4:730 weakly excites
  13. L12:2327 weakly excites
  14. L4:1256 weakly excites
  15. L16:1672 weakly excites
  16. L4:2647 weakly suppresses
  17. L16:2790 weakly excites
  18. L10:2905 weakly suppresses
  19. L16:2655 weakly suppresses
  20. L10:1452 weakly excites
  21. L7:2058 weakly excites
  22. L4:1349 weakly suppresses
  23. L11:1469 weakly excites
  24. L12:812 weakly suppresses

Feeds later layers

  1. L19:1177 excites
  2. L24:1811 excites
  3. L22:1413 excites
  4. L19:1684 excites
  5. L21:645 excites
  6. L19:1367 suppresses
  7. L22:892 weakly excites
  8. L21:1509 weakly excites
  9. L23:271 weakly suppresses
  10. L22:228 weakly excites
  11. L20:1828 weakly excites
  12. L21:1810 weakly excites
  13. L24:1895 weakly excites
  14. L25:929 weakly excites
  15. L19:376 weakly suppresses
  16. L19:2395 weakly excites
  17. L24:1263 weakly excites
  18. L20:554 weakly excites
  19. L21:2586 weakly suppresses
  20. L23:2602 weakly suppresses
  21. L19:357 weakly suppresses
  22. L23:1060 weakly excites
  23. L23:298 weakly excites
  24. L27:804 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 -9.76 and -8.981 values between -8.98 and -8.211 values between -7.43 and -6.661 values between -5.10 and -4.331 values between -4.33 and -3.552 values between -3.55 and -2.784 values between -2.78 and -2.003 values between -2.00 and -1.23319 values between -1.23 and -0.453282 values between -0.45 and 0.33155 values between 0.33 and 1.1057 values between 1.10 and 1.8824 values between 1.88 and 2.659 values between 2.65 and 3.4312 values between 3.43 and 4.204 values between 4.20 and 4.984 values between 4.98 and 5.757 values between 5.75 and 6.535 values between 6.53 and 7.313 values between 7.31 and 8.082 values between 8.08 and 8.861 values between 11.96 and 12.731 values between 12.73 and 13.513 values between 13.51 and 14.291 values between 16.61 and 17.391 values between 17.39 and 18.162 values between 18.16 and 18.942 values between 20.49 and 21.27offnegativepositiveoften

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. being
  2. 具有
  3. possessed
  4. ffic
  5. having
  6. 具备
  7. 他们是
  8. being
  9. 是一种
  10. 拥有

Suppresses

  1. сол
  2. WithContext
  3. 夙
  4. MLA
  5. _axes
  6. önce
  7. ipur
  8. менно
  9. BaseContext
  10. ণ