'Indian' before Ocean or subcontinent
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 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 reacts to the word "Indian" when it comes before "Ocean", "subcontinent" or a place name, such as "the Indian Ocean". Turning it up or down did not give a clear theme in the generated 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 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.
-
Rasikajy” occupied settlements along the coast. They were exploiting the natural resources of the rich hinterland and participating in the large trade network of the Indian Ocean.⏎New fieldwork research is ongoing to investigate the production of stone vessels and
-
Suez Canal⏎The Suez Canal connects the Mediterranean Sea (foreground left) with the Gulf of Suez and thereby the Red Sea and the Indian Ocean and this way shortens the sea passage from Europe to the Middle and Far
-
faculty position in Computer Science at Queen’s University Belfast, United Kingdom. He received his B.Tech from CUSAT and his M.Tech and PhD from Indian Institute of Technology Madras, all in Computer Science. His current research interests include
-
AnimalWised and discover the native animals of the Thar Desert.⏎The Thar Desert: Home to diverse species⏎Located in the northwestern region of the Indian subcontinent, between Pakistan and India, the Thar desert has minimal vegetation yet
-
Hawai‘i: Northwest Hawaiian⏎Islands, O‘ahu, Maui, and Hawai‘i Island.⏎Worldwide: Eastern Atlantic,⏎Caribbean, Indian and Pacific Oceans.⏎Before the introduction of⏎Acanthophora sp
-
Viswanathan said with increasing degradation of the earth, environmental geosciences assumed critical importance. The recent earthquakes in Sumatra and Kashmir and the tsunami in the Indian Ocean had created greater awareness about the importance of environmental geosciences among policy makers
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 add a feature to a web app that is full of small black tiles that look like a black tile URL URL URL URL URL URL URL URL URL URL URL URL URL URL URL URL URL URL URL URL URL URL URL URL URL
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
it is freed from the clutches of British rule. During the partition of Bengal and freedom by the indigenous movement in which Sir Henry Cotton Surendra Nath Banerjee and turned into a
-
not be a recovery at all, but just an L-shaped continuation of economic misery.)⏎But not everyone subscribes to this framework of thought. Anirvan Banerji, director of research at
-
haMbhAraveNa cha |⏎vrajasyotthApanaM chakrurghoShavR^iddhAH samAgatAH ||2-9<|endoftext|>A population of fruit bats which is found across much
-
Dakini⏎- The Buddhist protector deity Penden Lhamo⏎- The Goddess of the White Umbrella (the Buddhist deity Ushnisha-sitatapatra)⏎- The Great mystic Virupa⏎- Thunderbolt and bell⏎
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
- L23:397 strongly excites
- L20:500 strongly suppresses
- L19:1228 excites
- L10:190 excites
- L10:2008 suppresses
- L21:2222 suppresses
- L20:1027 excites
- L18:477 excites
- L9:2483 excites
- L18:177 suppresses
- L3:62 excites
- L16:692 suppresses
- L9:144 suppresses
- L18:560 suppresses
- L10:556 excites
- L4:1890 suppresses
- L9:429 excites
- L7:546 suppresses
- L8:495 suppresses
- L20:2345 excites
- L9:812 excites
- L9:2482 suppresses
- L11:840 suppresses
- L21:285 excites
Feeds later layers
- L27:2458 excites
- L25:1504 weakly excites
- L25:2687 weakly excites
- L25:1525 weakly excites
- L25:2761 weakly suppresses
- L27:847 weakly excites
- L26:2807 weakly excites
- L26:2004 weakly excites
- L25:2086 weakly suppresses
- L25:574 weakly excites
- L25:1804 weakly suppresses
- L25:254 weakly excites
- L25:2307 weakly suppresses
- L25:485 weakly excites
- L25:2725 weakly excites
- L25:2432 weakly excites
- L27:1001 weakly excites
- L26:1991 weakly excites
- L25:1078 weakly suppresses
- L25:2203 weakly excites
- L25:230 weakly suppresses
- L25:2795 weakly excites
- L25:2458 weakly excites
- L27:516 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
- _IND
- inds
- Ind
- IND
- _ind
- Indians
- Ind
- Gandhi
- IND
- _definition
Suppresses
- เกร
- reibung
- epub
- zo
- spots
- Appe
- 圩
- bug
- خلا
- ник