"late" in dated phrases like "in the late 1980s"
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. 5 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 "late" when it comes right after "in the" and before a decade or century, as in "in the late 1800s." When it was turned up, the model's text drifted toward unrelated rambling, so the steering did not show 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.
-
ash-spewing hell vent whose name only Icelanders could pronounce—the B.C. volcanologist might have been best equipped to rationalize her frustration. Back in the late 1980s, Hickson had helped spearhead efforts to create
-
".⏎Famed British author Norman Davies noted that American forces fought with distinction on two fronts, but that it was our home factories that turned the tide. In the late 1930's, American industry produced consumer goods at a dog's
-
permitted lamps to burn whale-oil and, later, kerosene—today, these are the primary kinds of antique lamps most favored by collectors. In the late 18th century, Swiss chemist Aime Argand invented the first
-
Federal authorities for her Southern nationalism. A younger brother, Samuel, was a major in the Georgia infantry and spent time as a Union prisoner of war.⏎In the late 1840s, the Levy family left Charleston for
-
explanation. As the last remaining fabric from a series of four, Oxford cloth’s story is one of inexplicable survival. In the late nineteenth century a Scottish Mill designed and produced four cloths, one for each of
-
political discussion and debate in digital space having on real life politics? How is the Internet changing actual politics?⏎When the Internet began arriving in the Middle East in the late 1990s, some had high expectations that it would usher in
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, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the uprising⏎of the British Empire. But we have no right to this position of special⏎and honorable responsibility unless we are determined to use it to the⏎full, not for our own
-
Should the money follow the people, who have flocked for decades to the Sun Belt, where homes are cheap and jobs are plentiful but climate risks mount by the day? Or, following the logic of Biden’s own order, should funds be directed
-
a goat wants to go somewhere, and you don’t have a good hold on her, she’ll drag you along for the ride. She’s usually heading for the nearest fruit<|endoftext|>In a few days a carrier rocket
-
each other to no avail. Each person’s actions were canceled out by the others’ actions. The sub just sat there.⏎Finally, from the back of the room, Carpenter’s voice suggested, “Why don’t you guys go
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
- L2:205 excites
- L3:299 weakly excites
- L3:344 weakly excites
- L3:2904 weakly excites
- L0:2693 weakly suppresses
- L0:1061 weakly suppresses
- L4:1984 weakly suppresses
- L4:2904 weakly excites
- L3:2998 weakly suppresses
- L2:2703 weakly suppresses
- L1:2239 weakly excites
- L1:2056 weakly suppresses
- L2:1439 weakly excites
- L4:455 weakly suppresses
- L4:12 weakly suppresses
- L3:2698 weakly suppresses
- L2:1880 weakly suppresses
- L0:1197 weakly excites
- L1:491 weakly excites
- L1:1244 weakly suppresses "of" after "board" or "off"
- L1:507 weakly excites
- L1:408 weakly excites
- L1:86 weakly suppresses
- L2:2388 weakly suppresses
Feeds later layers
- L6:2164 strongly excites second half of multi-word proper names
- L8:849 suppresses
- L10:457 suppresses
- L7:26 suppresses
- L6:2283 suppresses
- L9:716 excites
- L9:1607 suppresses
- L10:1316 excites
- L9:929 excites 'each' in 'each other'
- L14:1403 excites
- L16:662 suppresses
- L16:656 suppresses
- L10:217 suppresses
- L10:952 suppresses
- L6:1463 excites
- L15:1601 suppresses
- L8:98 excites
- L9:845 excites
- L7:101 excites
- L13:1550 suppresses
- L6:1354 excites
- L12:576 excites 'of' in fixed phrases like table of contents
- L7:111 excites
- L13:1513 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
- UMP
- ative
- 下去
- tastes
- umps
- ものが
- hon
- �
- ��
- ствие
Suppresses
- (Mat
- FixedSize
- -CS
- (Role
- >No
- pagesize
- arness
- (Job
- tyres
- -lat