'of' after geographic nouns
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. 12 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly 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 lights up on the word 'of' when it follows a place or landmark, like 'shores of Hawaii', 'floor of the mouth', or 'west coast of Mexico'. It is quiet on most other text. Turning it up gave repetitive sentences that were not clearly about geography.
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.
-
of the massive convergence of marine litter known as the Great Pacific Garbage Patch. The easternmost concentration of trash is midway between the California coast and the eastern shores of Hawaii.⏎These
-
the oral cavity is produced by the submandibular glands, even though they are much smaller than the .⏎The are a pair of glands located beneath the floor of the mouth anterior to the . The secretion produced is mainly in nature, however it
-
churban helps us mourn its destruction. Tisha B’Av is the culmination, of course, of two national tragedies: the loss of the first Beis HaMikdash as well as the Second Beis HaMikdash.⏎To
-
the worst. Learn more about Ninja Polymers in the infographic below, and share individual story sections by clicking on the embed icon in the upper right-hand corner of
-
shall begin on the Gulph [sic] of Mexico, at the mouth of the river Sabine, in the sea, continuing north, along the western bank of that river, to the 32d degree of latitude; thence,
-
way stretches 4,000 miles north-to-south and 1,000 miles east-to-west. From the Arctic to the west coast of Mexico and the Rocky Mountains to
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: [X]A and X]A, the same A, are [X]A (A)X (X) [A]X (A)X (A)A. (A)X (A)A (A
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
, fellow skeptic, and fellow blogger Jeff Medkeff discovered the asteroid in 2000. It was given the preliminary designation of 2000 WG11, and Jeff had the privilege of naming it, and the short
-
For further information and to get help call the Poison Help Line at 1-800-222-1222.⏎Synonyms and
-
010] Please see my latest post about setting an ad hoc network to share an internet connection.<|endoftext|>Encyclopedia of Southern Jewish Communities⏎History of Congregation B'nai Israel, Little Rock⏎B’nai Israel congregation was first organized
-
're saving.⏎For More Information...⏎Phone: 1-888-STAR-YES (1-888-782-7937)⏎For more information on activities ongoing
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
- L11:1246 excites
- L10:1748 excites
- L11:2953 suppresses
- L6:1563 excites
- L6:2664 excites
- L8:1544 excites
- L10:2786 excites
- L8:526 suppresses
- L8:2956 excites
- L8:502 excites
- L8:660 suppresses
- L9:1428 excites
- L10:2617 suppresses
- L8:1755 suppresses
- L10:1395 excites
- L8:1003 suppresses "the" or "of" before a noun phrase
- L6:1244 suppresses
- L11:1611 excites
- L6:1412 suppresses
- L5:2114 weakly suppresses unclear: calculus notation, mixed topics
- L10:1706 weakly suppresses
- L9:998 weakly suppresses
- L9:2764 weakly suppresses
- L10:272 weakly suppresses
Feeds later layers
- L14:1659 suppresses
- L13:856 suppresses
- L13:210 suppresses
- L16:3012 suppresses
- L20:331 excites
- L18:725 suppresses
- L18:370 suppresses
- L20:1112 suppresses
- L14:539 suppresses
- L16:229 suppresses
- L14:126 excites
- L13:2049 suppresses
- L20:1218 suppresses
- L16:1321 suppresses
- L19:414 suppresses
- L14:1971 suppresses
- L15:1299 suppresses
- L17:1240 excites
- L19:761 suppresses
- L14:2275 suppresses
- L14:655 suppresses
- L16:985 weakly suppresses
- L15:1973 weakly excites
- L17:264 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
- nock
- ัย
- forfeiture
- =?";⏎
- .blogspot
- uelle
- столь
- edException
- 煲
- such
Suppresses
- Mass
- anything
- Var
- entimes
- inos
- Mass
- ((*
- olate
- 奥
- ovo