Neuron Explorer

Qwen3-0.6B

unclear: 'the' inside 'end of' phrases

A neuron in layer 14 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 is hard to pin down. It fires on 'the' in ordinary phrases, like 'into the water' or 'of the Civil War', and it pushes strongly the other way on phrases like 'at the end of the' or 'leave the'. Turning it up or down produced no clear pattern in the writing.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    no problem hearing everything. When the bass and toms are added, conflicts begin to appear between the bass<|endoftext|>The Bruce-Benedicks Saga and the Birth of the AGH⏎While living in Paris in the 1880's,

  2. #2 peak

    should learn by the end of the second and third grade. Remember that these skills are not to be used in determining whether or not a child should be promoted to the next level but to get a better idea

  3. #3 peak

    plug the stem holding the water in then stand the flower upright in a vase of water.⏎- To increase the bloom time of cut flowers, plunge their stems into the water and let them stand overnight before arranging them. A crushed aspirin, a

  4. #4 peak

    collections to collaborate.⏎Organize your curriculum with collections. Easy!⏎Have time to be more creative & energetic with your students!⏎Students explore the Orange Revolution. In this Orange Revolution lesson, students examine the climate of the Ukrainian protest and investigate the role

  5. #5 peak

    @example.com.<|endoftext|>Table of Contents :⏎Top Suggestions Senior Kindergarten Writting Worksheet :⏎Senior Kindergarten Writting Worksheet In this worksheet your students will practice their handwriting skills by tracing the

  6. #6 peak

    on their solid oxide fuel cell, also known as a SOFC. The pilot system they built generates about 2 kW of electricity, or how much power a typical American home consumes. The PNNL team designed its system so it can be

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 think that the problem of the first order, the first order of the problem, I think that the problem of the first order, the first order, the problem, the answer, the question, the answer, it's the problem,

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

    - Real world examples of task models (from the travel, e-commerce, financial and automotive sectors), and the differences they've made to projects⏎Participants will leave the workshop armed with the ability to create their own task models and make a real difference

  2. #2 peak

    prepare and plan in conducting successful negotiations. It will also help participants gain greater insight into various negotiation issues currently involved in different WTO Agreements.⏎At the end of the course, the participants should be able to:⏎• Differentiate between bilateral and mult

  3. #3 peak

    least expects it. Paterson always establishes hope and optimism even in her most bleakly realistic stories, which dramatize internal and external conflicts. At the conclusion of the novel, Gilly still faces difficult choices, but she knows herself better and

  4. #4 peak

    explore some ways AI can help you create better content for your blog. We will cover everything from finding your niche to writing and editing your content. By the end of this post, you should understand how AI can help you take your content to the

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. L13:1405 suppresses
  2. L13:1100 excites
  3. L8:2755 suppresses
  4. L12:1906 weakly suppresses
  5. L13:1456 weakly excites
  6. L11:2037 weakly excites
  7. L4:968 weakly excites
  8. L8:1003 weakly suppresses "the" or "of" before a noun phrase
  9. L11:2497 weakly suppresses
  10. L7:790 weakly excites
  11. L9:2806 weakly suppresses
  12. L12:3004 weakly excites
  13. L13:956 weakly suppresses
  14. L13:210 weakly suppresses
  15. L12:2849 weakly excites '-based' after a place name
  16. L12:1065 weakly excites
  17. L6:1735 weakly excites
  18. L13:2594 weakly excites
  19. L9:1909 weakly excites
  20. L12:2176 weakly excites 'of' after a place or role noun
  21. L6:1218 weakly suppresses
  22. L12:2544 weakly excites
  23. L7:111 weakly suppresses
  24. L11:2207 weakly suppresses "of" after a noun in trademark and title phrases

Feeds later layers

  1. L19:1900 suppresses
  2. L18:1422 suppresses
  3. L16:371 suppresses
  4. L20:1983 suppresses
  5. L20:337 excites
  6. L20:714 suppresses
  7. L20:1233 suppresses
  8. L21:1119 suppresses
  9. L17:952 suppresses
  10. L21:368 suppresses
  11. L15:537 suppresses middle digits of numbers
  12. L16:938 suppresses
  13. L21:843 suppresses
  14. L20:254 weakly suppresses
  15. L18:77 weakly suppresses
  16. L21:309 weakly excites
  17. L17:265 weakly suppresses
  18. L16:2612 weakly suppresses
  19. L16:1643 weakly suppresses
  20. L19:638 weakly suppresses
  21. L16:2705 weakly suppresses
  22. L17:429 weakly excites
  23. L19:2609 weakly excites
  24. L19:932 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.

2 values between -6.68 and -6.451 values between -5.30 and -5.073 values between -4.38 and -4.151 values between -4.15 and -3.921 values between -3.92 and -3.691 values between -3.69 and -3.464 values between -3.46 and -3.232 values between -3.23 and -3.002 values between -3.00 and -2.773 values between -2.77 and -2.542 values between -2.54 and -2.314 values between -2.31 and -2.083 values between -2.08 and -1.8510 values between -1.85 and -1.6219 values between -1.62 and -1.3914 values between -1.39 and -1.1628 values between -1.16 and -0.9337 values between -0.93 and -0.7070 values between -0.70 and -0.47187 values between -0.47 and -0.241375 values between -0.24 and -0.011788 values between -0.01 and 0.22236 values between 0.22 and 0.4555 values between 0.45 and 0.6830 values between 0.68 and 0.9117 values between 0.91 and 1.147 values between 1.14 and 1.371 values between 1.37 and 1.601 values between 1.60 and 1.831 values between 1.83 and 2.062 values between 2.06 and 2.291 values between 2.29 and 2.52offnegativepositiveoften

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. desk
  2. unge
  3. ệt
  4. acht
  5. 胼
  6. ung
  7. дет
  8. ome
  9. urses
  10. erea

Suppresses

  1. >Last
  2. .Lookup
  3. 拿出来
  4. prestashop
  5. otti
  6. teardown
  7. 要说
  8. гон
  9. pulls
  10. .INSTANCE