Neuron Explorer

Qwen3-0.6B

"over" in "take over"

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 lights up on the word "over" in the phrase "take over", as in "robots can take over rote tasks". It also shows a weaker negative response to "begin" in "To begin with". Turning it up gave unrelated text, so the steering did not confirm the pattern.

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

    . Here are some great examples in a musical format.⏎We will be looking at numbers next time then give and take over the next two weeks for Halloween, sobe sure to let me know by visiting

  2. #2 peak

    feel like they don’t know where they’re going and lose sight of their goal⏎- let procrastination take over⏎So how do you help your teen learn how to create an action plan when

  3. #3 peak

    which can be found on Facebook) have been working on the latter solution and have made significant progress towards the creation of an automated tool, potentially allowing computers to take over. But now,<|endoftext|>In today’s virtual workshop, artist Sophie Daly will show

  4. #4 peak

    , and rest are subject to a constellation of taboos and prescriptions. Sex is a no-no. Rest is mandated and traditionally facilitated by female relatives who take over errands and chores. Foods are divided into the approved (carrots, chicken

  5. #5 peak

    . Most automated machinery and tools can operate at an optimal level, as well as improve working conditions for those spending the most time inside such facilities. Robots can take over rote tasks, freeing up human workers to handle more exciting duties. They can

  6. #6 peak

    7-Separated from Baylor University.⏎1910-Moved to Fort Worth.⏎1918-College of Dentistry organized in Dallas by taking over the State Dental College which had been founded in 1905.⏎1

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 need to be more precise. I have to think about this problem in my own way. So, let me think again about the problem. The problem is to find the minimal value of the function. So, first, let me try

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

    can the climate generation do?⏎To begin with, business as usual is not an option. These young people are green entrepreneurs

  2. #2 peak

    mediated mRNA isolation. Here, we describe a protocol that employs sample derived rRNA capture probes to remove rRNA from a metagenomic total RNA sample. To begin, both mosquito and microbial small and large subunit rRNA fragments are amplified from

  3. #3 peak

    to take a stab at it. Put on your seatbelt, because wrapping your head around TVOC can be a bit of a mental rollercoaster!⏎To begin our TVOC journey, let’s start off with a simple question: What is

  4. #4 peak

    advantages. Take a look.⏎Without a doubt, swept-source OCT has positively impacted and exceptionally contributed to

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. L0:178 weakly suppresses
  2. L0:1301 weakly suppresses
  3. L1:1955 weakly excites
  4. L0:1224 weakly suppresses
  5. L0:2835 weakly excites
  6. L1:2907 weakly suppresses
  7. L1:3032 weakly suppresses
  8. L1:90 weakly excites
  9. L0:1962 weakly excites
  10. L0:659 weakly suppresses
  11. L1:1404 weakly suppresses
  12. L1:2318 weakly suppresses
  13. L1:379 weakly excites
  14. L1:1692 weakly excites
  15. L1:2197 weakly suppresses
  16. L1:2610 weakly excites
  17. L0:1204 weakly suppresses
  18. L1:2653 weakly suppresses
  19. L0:2348 weakly excites
  20. L1:1584 weakly suppresses
  21. L0:2952 weakly excites
  22. L1:559 weakly suppresses
  23. L0:48 weakly excites
  24. L0:1723 weakly excites

Feeds later layers

  1. L4:117 suppresses
  2. L9:2276 weakly excites
  3. L17:2499 weakly suppresses
  4. L14:1818 weakly excites
  5. L20:2721 weakly suppresses
  6. L3:1888 weakly excites
  7. L3:2293 weakly excites
  8. L14:2167 weakly suppresses
  9. L10:2664 weakly suppresses
  10. L9:1730 weakly suppresses
  11. L3:1661 weakly suppresses
  12. L3:1660 weakly excites 'ago' after a span of years
  13. L23:230 weakly suppresses
  14. L19:1224 weakly suppresses
  15. L4:2058 weakly suppresses
  16. L20:1995 weakly suppresses
  17. L4:825 weakly excites
  18. L10:459 weakly excites
  19. L7:1554 weakly suppresses
  20. L13:2964 weakly suppresses
  21. L20:2315 weakly suppresses
  22. L9:1919 weakly excites
  23. L4:360 weakly suppresses
  24. L12:2325 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 -2.84 and -2.751 values between -2.75 and -2.651 values between -1.07 and -0.981 values between -0.98 and -0.882 values between -0.79 and -0.703 values between -0.70 and -0.612 values between -0.61 and -0.514 values between -0.51 and -0.4216 values between -0.42 and -0.3349 values between -0.33 and -0.23182 values between -0.23 and -0.14676 values between -0.14 and -0.051790 values between -0.05 and 0.05886 values between 0.05 and 0.14237 values between 0.14 and 0.2345 values between 0.23 and 0.335 values between 0.33 and 0.423 values between 0.42 and 0.511 values between 0.61 and 0.701 values between 0.70 and 0.792 values between 0.79 and 0.89offnegativepositiveoften

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. _CN
  2. ariat
  3. 扇
  4. ult
  5. 丞
  6. ork
  7. 增量
  8. 价
  9. 侄
  10. enstein

Suppresses

  1. raud
  2. lesen
  3. aylight
  4. explanation
  5. материал
  6. .AI
  7. ASON
  8. truth
  9. :⏎⏎
  10. prites