noun right after a possessive
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 1 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron responds to the word that comes right after 'my', 'her', 'his' or 'their', such as 'readers' in 'my readers' or 'audience' in 'his audience'. It is weaker and sometimes negative on places or people after possessives like 'my apartment'. Turning it up did not produce a clear theme.
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 activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
my readers will be surprised or even upset that Thornton puts profit at the top of her list
-
20 Images Online|⏎(48 votes cast)⏎Biggles and the team set out as part of the International Squadron to help the Finns. They stumble upon a dying Pole and become involved in a tussle with German intelligence for his
-
is kind energy coming from her. I feel light and happy, and sometimes I even feel like I’ve worked through something while in her presence.⏎I’ve experienced it during my years of practicing optometry.⏎Holding space
-
.com. Sadly they wanted someone with more experience but hey I tried right? So instead of deleting my short article I thought I would share it with my readers. I cannot use it for school purposes since I have to only use material found
-
2005 President George Bush delivered a speech designed to justify his foreign policy failures since 9/11. Throughout his talk he intoned his audience to recognize the goodness of American foreign policy and the inherent democratic character of the United
-
. As you write her words, bring her attention to the use of uppercase and lowercase letters and spacing between words.⏎“C-a
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
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.
Here is something I have been thinking about lately: I want to be able to do this, and I have been thinking about it a lot. I have been thinking about it again, and I have been thinking about it again again, and I have been thinking about it again again again,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
"As the alerts came in, my mind began to race," Jenna Wortham wrote. "Three friends, I learned, had arrived at a music venue near my apartment. But why? What was happening there? Then I saw pictures of other friends
-
EMPLOYED HERE". In the early hours of the morning, as the rioting spread north and south, Lino Rivera was picked up from his mother's apartment and photographed with a police officer. The photographs were distributed in order to prove that
-
other personal details. This also includes the name of your school, your photo or any personal information about your friends or family.⏎- Meeting people online might be fun, but remember the people you meet online
-
stays local. Users are able to communicate with their immediate surrounding communities. Neighbors can get to know each other, discuss local issues or even search for a babysitter. It was a conversation on Nextdoor that prompted us to start Tech For Non
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L0:2789 weakly suppresses
- L0:1760 weakly excites
- L0:2333 weakly suppresses
- L0:1223 weakly excites
- L0:2107 weakly excites
- L0:3040 weakly suppresses
- L0:393 weakly excites
- L0:1285 weakly suppresses
- L0:1980 weakly excites
- L0:685 weakly excites
- L0:386 weakly suppresses
- L0:1180 weakly excites
- L0:40 weakly excites 'of' in 'sort of' hedges
- L0:3044 weakly excites
- L0:1695 weakly excites
- L0:362 weakly suppresses
- L0:1823 weakly excites
- L0:1986 weakly excites
- L0:1854 weakly excites
- L0:2091 weakly excites
- L0:1433 weakly excites
- L0:895 weakly excites
- L0:2642 weakly excites
- L0:677 weakly suppresses
Feeds later layers
- L16:686 weakly excites
- L4:726 weakly suppresses
- L13:173 weakly suppresses
- L7:1486 weakly suppresses
- L16:1885 weakly excites
- L18:628 weakly suppresses
- L12:488 weakly excites
- L25:2563 weakly excites
- L12:271 weakly excites
- L27:2419 weakly excites
- L11:1318 weakly excites
- L17:1318 weakly excites
- L24:1064 weakly suppresses
- L12:90 weakly excites
- L17:2575 weakly suppresses
- L13:1066 weakly suppresses
- L16:51 weakly suppresses
- L7:2072 weakly suppresses
- L9:2681 weakly suppresses
- L18:1081 weakly suppresses
- L4:2504 weakly suppresses
- L2:390 weakly excites
- L10:3002 weakly excites
- L5:981 weakly suppresses
Activation histogramA chart of how often a neuron takes each activation value.
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.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
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 layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- .coroutines
- olduğunu
- aturas
- врем
- -President
- PropertyDescriptor
- reative
- .Properties
- +xml
- Hast
Suppresses
- 술
- Lana
- abled
- pak
- bit
- ngr
- mits
- oken
- avax
- itel