Neuron Explorer

Qwen3-0.6B

'that' starting a clause after a belief

A neuron in layer 9 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

The neuron lights up on the word 'that' when it introduces a belief or claim, like 'a misconception is that' or 'people think that'. It goes quiet, or strongly negative, on phrases like 'in fact', which mark a correction. Turning it up did not produce any clear theme in the sample text.

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

    than the general population to suffer kidney failure later in life, probably in part because they were so super-healthy to start.⏎'There is this perception out there that donors don't want to be followed up. That's not necessarily

  2. #2 peak

    news. Depression is highly treatable. Studies show that a combination of antidepressant medication and psychotherapy is the best prescription.⏎One very common misconception about depression is that it feels like mere sadness. Sadness is only a part of it. It

  3. #3 peak

    intent to mislead in order to gain financially or politically.⏎Fake News vs Satire⏎If a well known comedy programme featured a parody of a news presenter saying that Donald Trump’s hair was actually a miniaturised Siberian Yeti implanted by

  4. #4 peak

    salt-water marine locales. I often hear that fresh water boats don’t even need anodes, or that zinc anodes are fine because they last forever, or that boaters shouldn’t mix anode metals with different metals underwater. Wrong, wrong

  5. #5 peak

    to get professional help.⏎Fathers, too, are profoundly affected by the loss of a child. Unfortunately, a common misconception regarding miscarriage and stillbirth is that only the mother is affected. Women often feel more freedom to cry and express their

  6. #6 peak

    don't remember "not" well. If you try to sell people on the concept that "Barack Obama is a Christian," you might convince some of them that he's not a Muslim. But if you try to sell people on "Bar

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 don't have to look for the problem. The problem is that I'm looking for the solution. The solution is that there is no solution. There is also a solution when the problem is not the case. That is, there is

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

    ANDAS can be challenging even for the most skilled clinicians. Far too often, children and adolescents are misdiagnosed as having a mental illness, when in fact, they have an autoimmune disease, triggered by an infection.⏎Cunningham Panel

  2. #2 peak

    include feeling bloated, uncomfortable, and sluggish. In most cases, constipation is temporary and not serious.⏎Many people think they are constipated when, in fact, their bowel movements are regular. For example, some people believe they are const

  3. #3 peak

    avoid the mistake, you need to understand well what kind of fertile soil. Many people have the thought that the color of fertile soil is brown or red. In fact, black fertile soil is suitable for plants with flowers, while the brown or red

  4. #4 peak

    reemergence of whooping cough. Many people erroneously believe tetanus can only be acquired from rusty nails. In actuality, tetanus is all around us and can contaminate any skin wound.

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. L8:811 suppresses
  2. L7:2109 weakly excites
  3. L6:431 weakly excites
  4. L8:2677 weakly suppresses
  5. L2:678 weakly excites
  6. L2:1004 weakly excites
  7. L8:184 weakly excites
  8. L7:384 weakly excites
  9. L1:1155 weakly excites
  10. L3:2023 weakly suppresses
  11. L8:13 weakly suppresses
  12. L5:358 weakly suppresses
  13. L7:2488 weakly excites
  14. L7:71 weakly suppresses
  15. L8:421 weakly suppresses
  16. L5:1116 weakly excites
  17. L6:2447 weakly suppresses
  18. L8:2336 weakly suppresses
  19. L3:1907 weakly excites
  20. L4:2891 weakly suppresses
  21. L1:1564 weakly suppresses
  22. L4:3038 weakly suppresses
  23. L7:1095 weakly excites
  24. L0:2779 weakly suppresses

Feeds later layers

  1. L12:2574 suppresses
  2. L18:2248 suppresses
  3. L15:2891 suppresses
  4. L13:1055 suppresses
  5. L13:2930 weakly suppresses
  6. L17:2752 weakly excites
  7. L11:1065 weakly suppresses
  8. L21:1481 weakly suppresses 'the' before a topic noun
  9. L18:2619 weakly excites
  10. L10:251 weakly suppresses
  11. L25:2109 weakly excites
  12. L12:3069 weakly excites
  13. L19:2330 weakly excites
  14. L11:922 weakly excites
  15. L15:242 weakly excites
  16. L12:1987 weakly suppresses
  17. L11:2224 weakly excites
  18. L12:2647 weakly suppresses
  19. L20:2182 weakly suppresses
  20. L14:1305 weakly excites
  21. L13:3048 weakly suppresses
  22. L25:2212 weakly excites
  23. L11:71 weakly suppresses
  24. L12:1918 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 -5.49 and -5.331 values between -1.56 and -1.401 values between -1.40 and -1.241 values between -1.24 and -1.078 values between -0.75 and -0.5841 values between -0.58 and -0.42207 values between -0.42 and -0.26840 values between -0.26 and -0.091714 values between -0.09 and 0.07812 values between 0.07 and 0.23228 values between 0.23 and 0.4043 values between 0.40 and 0.567 values between 0.56 and 0.733 values between 0.73 and 0.891 values between 0.89 and 1.05offnegativepositiveoften

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. anches
  2. FullYear
  3. .record
  4. Followers
  5. overn
  6. 紀
  7. .fill
  8. .oracle
  9. frau
  10. meno

Suppresses

  1. l
  2. ogn
  3. disturbing
  4. ises
  5. ovid
  6. ane
  7. <?,
  8. omet
  9. A
  10. entertaining