word fragment starting a longer name
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. 24 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 fires on the first syllables of a longer word, especially names like 'Zo' in Zoanthidae or 'Og' in Ogden. Turning it up made the model repeat the word 'over' again and again.
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.
-
colonial anthozoans and soft coral by their characteristic of incorporating sand and other small pieces of material into their tissue to help make their structure (except for the family Zoanthidae). A main characteristic of the order is that their tentacles are
-
ite blog about Israeli museums’ spotlight on Circassians.⏎“We’ve traditionally been friends with both Jews and Arabs, even when they were fighting among themselves,” said Zoher Thawcho, the 39-year-old co-founder of the Circ
-
by total area.⏎The county extends from high in the Wasatch Range in the east into a portion of the Great Salt Lake to the west. The Weber and Ogden rivers and their tributaries run through its valleys. The Weber County Survey
-
this was an easy matter and a parergon. They must apply themselves zealously and in the deepest secrecy to the study
-
anthids are the zooxanthellate genera found in tropical and sub-tropical waters (primarily Zoanthus and Palythoa), many other species and genera exist, some
-
B" and "C" Batteries⏎saw plenty of action and subdued several machine gun nests that were holding⏎up the infantry, whose initial pace from Egret trench
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: The main source of each person from the r class has to be from the r class each person each r class each person each person from the r class each r class each person each r class each person each r class each person each r class
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
data. Further, the cloud and artificial intelligence have a symbiotic relationship. The cloud handles the fundamental source of information that augments artificial intelligence networks, and artificial<|endoftext|>If I were to think of an emblem
-
finding, the researchers have clarified the way in which this oscillator adjusts its activity throughout the day. They show that the three proteins studied, the Pseudo-Response Regulators PRR5, PRR7 and PRR9, associate with promoter
-
. To the region’s most ardent segregationists, the Douglas DC-4s and Lockheed Electras landing from faraway cities on shimmering southern runways augured a<|endoftext|>Canada's tar sands, located in the province of Alberta, are
-
Bhushan, who inaugurated the conference, said the philosophy of sustainable development required maintaining a balance between the impact of anthropogenic activities and the ecosystem.⏎Inaugurating the conference, Anna University Vice-Chancellor
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
- L23:2981 suppresses
- L2:1632 weakly suppresses
- L22:384 weakly excites
- L6:2398 weakly excites
- L23:1489 weakly excites
- L19:1004 weakly suppresses
- L22:715 weakly suppresses
- L15:2362 weakly excites
- L2:1774 weakly suppresses
- L8:1506 weakly suppresses
- L17:360 weakly excites
- L9:923 weakly excites
- L8:418 weakly excites
- L11:1031 weakly excites
- L2:1229 weakly excites
- L19:2271 weakly excites
- L18:1350 weakly suppresses
- L9:213 weakly excites
- L16:1236 weakly suppresses
- L19:1052 weakly excites
- L2:2255 weakly suppresses
- L13:1820 weakly excites
- L18:2383 weakly excites
- L15:2116 weakly excites
Feeds later layers
- L26:911 weakly excites
- L25:1686 weakly suppresses
- L26:1505 weakly excites
- L27:75 weakly suppresses
- L25:126 weakly excites
- L26:2131 weakly suppresses
- L27:2251 weakly suppresses
- L27:285 weakly suppresses
- L26:2949 weakly excites
- L27:1977 weakly suppresses
- L27:1044 weakly excites
- L25:332 weakly excites
- L25:2625 weakly suppresses
- L25:619 weakly suppresses unclear: word fragments and 'or'/'and' in lists
- L27:3041 weakly suppresses
- L25:1256 weakly excites power and energy unit symbols
- L25:1468 weakly excites
- L27:945 weakly suppresses
- L27:1780 weakly suppresses
- L27:1294 weakly excites
- L25:402 weakly excites
- L27:455 weakly excites
- L26:2552 weakly excites
- L25:1788 weakly excites
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
- 过
- -over
- over
- over
- 過
- Over
- Over
- .over
- _over
- ancement
Suppresses
- ASF
- utzer
- [pos
- lus
- hazırl
- imshow
- 非要
- asper
- Manor
- ivr