Neuron Explorer

Qwen3-0.6B

comma after a leading 'Yes'

A neuron in layer 5 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 comma that follows the word 'Yes' at the start of an answer, as in 'Yes, goats can eat strawberries'. Turning it up did not make the output start with yes; the text drifted into unrelated material.

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

    interest of their constituents and of the nation. Now information technology has stripped away the insulation that physical distance provided back when information couldn’t travel faster than a horse.Yes, one wonders what the Framers might have thought of free speech.⏎February 3

  2. #2 peak

    turn, they will be kind to others. It is a amazing what a good message these gestures will send to them.⏎Teach The Value of Money⏎Yes, kids see you passing these green pieces of paper around, but do they really know

  3. #3 peak

    (at no extra cost to you). As an Amazon Associate I earn from qualifying purchases. Please read my disclaimer for additional details.⏎Can goats eat strawberries?⏎Yes, goats can eat strawberries, however, not all goats are fond of their sour and

  4. #4 peak

    welding techniques for a variety of purposes including hull repairs, welding anode brackets (see RNLI news story), repairing sheet pile walls and shuttering.<|endoftext|>Yes, there is blue light being emitted from your TV screen while you are watching TV.

  5. #5 peak

    them the right way, in the right time frame.⏎Is mixing baking powder and water a chemical change?⏎Is mixing water and baking soda a chemical change? Yes, mixing water and baking soda is a chemical change because it changes the composition of the

  6. #6 peak

    be responsible in school. Once you learn how to use the program, you can type or print lists and notices for yourself to be responsible in school.⏎- Yes, really a planner – A planner could actually help. The kind with a calendar and

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 first step in this blog is to set the stage for the problem by ensuring that all the devices in my network are connected to a private and public IP address. The private IP address is the main device of the network. The public IP

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

    Terraform is able to determine what changed and create incremental execution plans, which can be applied.⏎The infrastructure Terraform can manage includes low-level components such as compute instances, storage, and networking, as well as high-level components such as DNS entries

  2. #2 peak

    company could generate revenue from the new infrastructure investment. The first AWS service launched for public usage was Simple Queue Service in November 2004. Amazon EC2 was built by a team in Cape Town, South Africa, under Pinkham and

  3. #3 peak

    UGHTER THERAPY⏎Laughing your heart out can boost your immunity along with your mood. It raise the level of antibodies in the blood and wbc that kill viruses with increasing the number of antibodies in respiratory passages, the entry points

  4. #4 peak

    a form of inherited red blood cell disorder. The red blood cells consist of hemoglobin, a protein that carries oxygen to the tissues of the body. Healthy red blood cells are glossy, disk-shaped, and flexible.⏎Haemoglobin would be abnormal

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. L4:2865 suppresses
  2. L4:1869 suppresses
  3. L3:1220 weakly suppresses
  4. L4:2733 weakly excites
  5. L4:2312 weakly suppresses
  6. L2:1592 weakly excites
  7. L4:2301 weakly suppresses
  8. L2:1174 weakly suppresses
  9. L0:1715 weakly excites
  10. L3:1048 weakly excites
  11. L1:2644 weakly suppresses
  12. L0:595 weakly excites
  13. L2:2550 weakly excites
  14. L0:155 weakly excites
  15. L2:129 weakly excites
  16. L3:2421 weakly excites
  17. L4:1556 weakly suppresses
  18. L4:1299 weakly suppresses
  19. L0:177 weakly suppresses
  20. L4:2709 weakly suppresses
  21. L4:2557 weakly suppresses
  22. L2:2581 weakly excites
  23. L2:382 weakly excites
  24. L2:2192 weakly suppresses

Feeds later layers

  1. L13:1054 weakly suppresses
  2. L10:1958 weakly suppresses
  3. L6:1859 weakly suppresses
  4. L9:1991 weakly excites
  5. L17:1791 weakly excites
  6. L11:1514 weakly suppresses
  7. L10:2706 weakly excites
  8. L22:2375 weakly excites
  9. L20:2828 weakly excites
  10. L21:771 weakly suppresses
  11. L23:2156 weakly excites
  12. L22:2664 weakly excites
  13. L21:1521 weakly excites
  14. L14:2518 weakly excites
  15. L7:1206 weakly suppresses
  16. L17:2986 weakly excites
  17. L19:2314 weakly excites
  18. L23:1738 weakly excites
  19. L22:2017 weakly suppresses
  20. L6:1779 weakly excites
  21. L7:2283 weakly suppresses
  22. L16:1945 weakly suppresses
  23. L7:2800 weakly suppresses
  24. L23:967 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 -3.54 and -3.381 values between -1.08 and -0.912 values between -0.91 and -0.752 values between -0.75 and -0.5910 values between -0.59 and -0.42109 values between -0.42 and -0.26744 values between -0.26 and -0.092166 values between -0.09 and 0.07769 values between 0.07 and 0.2487 values between 0.24 and 0.4010 values between 0.40 and 0.561 values between 0.56 and 0.731 values between 0.73 and 0.891 values between 1.06 and 1.222 values between 1.55 and 1.712 values between 2.86 and 3.03offnegativepositiveoften

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. 掖
  2. ocument
  3. posé
  4. 钵
  5. enberg
  6. 一个职业
  7. hypocrisy
  8. eded
  9. eway
  10. heets

Suppresses

  1. ⌋
  2. тир
  3. ://${
  4. Tiles
  5. letal
  6. �
  7. below
  8. 指纹
  9. 】,【
  10. special