Neuron Explorer

Qwen3-0.6B

"the" starting an object noun phrase

A neuron in layer 20 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" right after a verb or preposition, such as "limits on the amount" or "access the site", where the next word will be the object of the sentence. Turning it up made the model repeat "the one that was", and turning it down gave broken text with "seen" and "common".

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

    to lend and how much interest to charge. By not knowing their credit score, people risk paying more interest when they take out a loan and may face limits on the amount they can borrow. In extreme circumstances, they may find themselves “locked out

  2. #2 peak

    types of fish. To put things into perspective, catfish love raw chicken liver while beam fish prefer insects. So arm yourself with a multitude of baits for the right occasion. Also if you have read about the fishing spot you’re going to

  3. #3 peak

    other pertinent information. Analyze additional psychopharmacological treatment options for bipolar disorders. Select a different anticonvulsant/mood-stabilizing medication than the one you previously selected (other than lithium) that a psychiatrist might prescribe to treat

  4. #4 peak

    backups protect healthcare providers from this problem.⏎Following Katrina, we saw the effects of physicians not having access to medical records because they were destroyed. Patients struggled to get the treatments and

  5. #5 peak

    to this problem that works reasonably well. I believe love is something we build. When I help others learn, I start by helping them build much strong love for the object through appreciating its beauty, connecting

  6. #6 peak

    attention to the people, places and events that you want to be part of your life, of your reality. It's simply a question of making a choice about the connections. Remembering at all times the choice is ours to make, we can

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 my my, my my, my my, my my, my my. My my my my. My my my my my. My my my my. My my my my. My my my my. I seen my seen my seen

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

    the productivity of cast aluminum, are widely used in the production of semi-continuous casting and die castings.⏎Foam Ceramic Filter Manufacturing Process⏎The main raw material of the porous filter material is high-temperature alumina. It does not react

  2. #2 peak

    how much more power you have solely due to<|endoftext|>Above: In the example above, students created a modular speaker set composed of interlocking triangular elements. The main materials were plywood veneer and cast plaster.⏎Above: After lengthy experimentation, the team

  3. #3 peak

    alin produced in Mexico. The patented Laetrile is a partly synthetic (man-made) form of amygdalin, while the laetrile/amygdalin made in Mexico comes from crushed apricot pits. The studies discussed in this fact

  4. #4 peak

    nose, and other parts too.We can also use buds to remove dust from the corners of windows, picture frames and statute We know everybody uses it. The main stuff for buds is cotton. So cigarette butts can be used as stuff for ear

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. L18:2575 strongly excites
  2. L19:2621 suppresses
  3. L17:2696 excites
  4. L17:2783 suppresses
  5. L18:1824 suppresses
  6. L18:414 excites
  7. L19:1235 suppresses
  8. L18:2740 suppresses
  9. L19:2179 excites
  10. L18:2364 excites
  11. L16:1785 excites
  12. L19:2542 suppresses
  13. L17:1811 excites
  14. L18:2666 suppresses
  15. L15:1851 suppresses 'that' after a noun, 'being' in passives
  16. L17:2240 excites
  17. L18:2378 suppresses
  18. L14:2528 suppresses
  19. L12:2519 suppresses
  20. L18:780 suppresses
  21. L19:409 excites
  22. L18:1267 excites
  23. L16:2746 suppresses
  24. L19:1185 excites

Feeds later layers

  1. L26:507 excites
  2. L22:1589 excites
  3. L23:94 suppresses
  4. L27:127 suppresses
  5. L21:914 suppresses
  6. L22:1118 suppresses
  7. L23:1024 suppresses
  8. L22:1124 suppresses
  9. L24:422 suppresses
  10. L21:1422 suppresses
  11. L21:191 excites
  12. L24:1366 suppresses
  13. L27:399 excites
  14. L24:484 suppresses
  15. L21:1851 suppresses
  16. L25:625 suppresses
  17. L21:863 weakly excites
  18. L27:217 weakly suppresses
  19. L23:360 weakly excites
  20. L22:1381 weakly suppresses
  21. L21:2413 weakly suppresses
  22. L21:2325 weakly excites
  23. L27:2875 weakly suppresses
  24. L27:1102 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.

1 values between -7.71 and -7.101 values between -6.49 and -5.871 values between -5.87 and -5.261 values between -5.26 and -4.651 values between -4.65 and -4.044 values between -4.04 and -3.436 values between -3.43 and -2.8212 values between -2.82 and -2.2134 values between -2.21 and -1.60147 values between -1.60 and -0.991001 values between -0.99 and -0.382007 values between -0.38 and 0.24617 values between 0.24 and 0.8541 values between 0.85 and 1.4613 values between 1.46 and 2.076 values between 2.07 and 2.684 values between 2.68 and 3.291 values between 3.29 and 3.905 values between 5.13 and 5.741 values between 5.74 and 6.351 values between 6.35 and 6.961 values between 6.96 and 7.571 values between 7.57 and 8.181 values between 16.12 and 16.73offnegativepositiveoften

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. the
  2. ones
  3. ethe
  4. .SetString
  5. なもの
  6. 类型的
  7. な
  8. The
  9. のに
  10. >The

Suppresses

  1. used
  2. seen
  3. used
  4. heard
  5. employed
  6. presented
  7. _seen
  8. swallowed
  9. listed
  10. ailability